fs/nfsd/nfs4state.c

   1 /*
   2 *  Copyright (c) 2001 The Regents of the University of Michigan.
   3 *  All rights reserved.
   4 *
   5 *  Kendrick Smith <kmsmith@umich.edu>
   6 *  Andy Adamson <kandros@umich.edu>
   7 *
   8 *  Redistribution and use in source and binary forms, with or without
   9 *  modification, are permitted provided that the following conditions
  10 *  are met:
  11 *
  12 *  1. Redistributions of source code must retain the above copyright
  13 *     notice, this list of conditions and the following disclaimer.
  14 *  2. Redistributions in binary form must reproduce the above copyright
  15 *     notice, this list of conditions and the following disclaimer in the
  16 *     documentation and/or other materials provided with the distribution.
  17 *  3. Neither the name of the University nor the names of its
  18 *     contributors may be used to endorse or promote products derived
  19 *     from this software without specific prior written permission.
  20 *
  21 *  THIS SOFTWARE IS PROVIDED ``AS IS'' AND ANY EXPRESS OR IMPLIED
  22 *  WARRANTIES, INCLUDING, BUT NOT LIMITED TO, THE IMPLIED WARRANTIES OF
  23 *  MERCHANTABILITY AND FITNESS FOR A PARTICULAR PURPOSE ARE
  24 *  DISCLAIMED. IN NO EVENT SHALL THE REGENTS OR CONTRIBUTORS BE LIABLE
  25 *  FOR ANY DIRECT, INDIRECT, INCIDENTAL, SPECIAL, EXEMPLARY, OR
  26 *  CONSEQUENTIAL DAMAGES (INCLUDING, BUT NOT LIMITED TO, PROCUREMENT OF
  27 *  SUBSTITUTE GOODS OR SERVICES; LOSS OF USE, DATA, OR PROFITS; OR
  28 *  BUSINESS INTERRUPTION) HOWEVER CAUSED AND ON ANY THEORY OF
  29 *  LIABILITY, WHETHER IN CONTRACT, STRICT LIABILITY, OR TORT (INCLUDING
  30 *  NEGLIGENCE OR OTHERWISE) ARISING IN ANY WAY OUT OF THE USE OF THIS
  31 *  SOFTWARE, EVEN IF ADVISED OF THE POSSIBILITY OF SUCH DAMAGE.
  32 *
  33 */
  34
  35 #include <linux/file.h>
  36 #include <linux/fs.h>
  37 #include <linux/slab.h>
  38 #include <linux/namei.h>
  39 #include <linux/swap.h>
  40 #include <linux/pagemap.h>
  41 #include <linux/ratelimit.h>
  42 #include <linux/sunrpc/svcauth_gss.h>
  43 #include <linux/sunrpc/addr.h>
  44 #include <linux/jhash.h>
  45 #include "xdr4.h"
  46 #include "xdr4cb.h"
  47 #include "vfs.h"
  48 #include "current_stateid.h"
  49
  50 #include "netns.h"
  51 #include "pnfs.h"
  52
  53 #define NFSDDBG_FACILITY                NFSDDBG_PROC
  54
  55 #define all_ones {{~0,~0},~0}
  56 static const stateid_t one_stateid = {
  57         .si_generation = ~0,
  58         .si_opaque = all_ones,
  59 };
  60 static const stateid_t zero_stateid = {
  61         /* all fields zero */
  62 };
  63 static const stateid_t currentstateid = {
  64         .si_generation = 1,
  65 };
  66 static const stateid_t close_stateid = {
  67         .si_generation = 0xffffffffU,
  68 };
  69
  70 static u64 current_sessionid = 1;
  71
  72 #define ZERO_STATEID(stateid) (!memcmp((stateid), &zero_stateid, sizeof(stateid_t)))
  73 #define ONE_STATEID(stateid)  (!memcmp((stateid), &one_stateid, sizeof(stateid_t)))
  74 #define CURRENT_STATEID(stateid) (!memcmp((stateid), &currentstateid, sizeof(stateid_t)))
  75 #define CLOSE_STATEID(stateid)  (!memcmp((stateid), &close_stateid, sizeof(stateid_t)))
  76
  77 /* forward declarations */
  78 static bool check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner);
  79 static void nfs4_free_ol_stateid(struct nfs4_stid *stid);
  80
  81 /* Locking: */
  82
  83 /*
  84  * Currently used for the del_recall_lru and file hash table.  In an
  85  * effort to decrease the scope of the client_mutex, this spinlock may
  86  * eventually cover more:
  87  */
  88 static DEFINE_SPINLOCK(state_lock);
  89
  90 enum nfsd4_st_mutex_lock_subclass {
  91         OPEN_STATEID_MUTEX = 0,
  92         LOCK_STATEID_MUTEX = 1,
  93 };
  94
  95 /*
  96  * A waitqueue for all in-progress 4.0 CLOSE operations that are waiting for
  97  * the refcount on the open stateid to drop.
  98  */
  99 static DECLARE_WAIT_QUEUE_HEAD(close_wq);
 100
 101 static struct kmem_cache *openowner_slab;
 102 static struct kmem_cache *lockowner_slab;
 103 static struct kmem_cache *file_slab;
 104 static struct kmem_cache *stateid_slab;
 105 static struct kmem_cache *deleg_slab;
 106 static struct kmem_cache *odstate_slab;
 107
 108 static void free_session(struct nfsd4_session *);
 109
 110 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops;
 111 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops;
 112
 113 static bool is_session_dead(struct nfsd4_session *ses)
 114 {
 115         return ses->se_flags & NFS4_SESSION_DEAD;
 116 }
 117
 118 static __be32 mark_session_dead_locked(struct nfsd4_session *ses, int ref_held_by_me)
 119 {
 120         if (atomic_read(&ses->se_ref) > ref_held_by_me)
 121                 return nfserr_jukebox;
 122         ses->se_flags |= NFS4_SESSION_DEAD;
 123         return nfs_ok;
 124 }
 125
 126 static bool is_client_expired(struct nfs4_client *clp)
 127 {
 128         return clp->cl_time == 0;
 129 }
 130
 131 static __be32 get_client_locked(struct nfs4_client *clp)
 132 {
 133         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 134
 135         lockdep_assert_held(&nn->client_lock);
 136
 137         if (is_client_expired(clp))
 138                 return nfserr_expired;
 139         atomic_inc(&clp->cl_refcount);
 140         return nfs_ok;
 141 }
 142
 143 /* must be called under the client_lock */
 144 static inline void
 145 renew_client_locked(struct nfs4_client *clp)
 146 {
 147         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 148
 149         if (is_client_expired(clp)) {
 150                 WARN_ON(1);
 151                 printk("%s: client (clientid %08x/%08x) already expired\n",
 152                         __func__,
 153                         clp->cl_clientid.cl_boot,
 154                         clp->cl_clientid.cl_id);
 155                 return;
 156         }
 157
 158         dprintk("renewing client (clientid %08x/%08x)\n",
 159                         clp->cl_clientid.cl_boot,
 160                         clp->cl_clientid.cl_id);
 161         list_move_tail(&clp->cl_lru, &nn->client_lru);
 162         clp->cl_time = get_seconds();
 163 }
 164
 165 static void put_client_renew_locked(struct nfs4_client *clp)
 166 {
 167         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 168
 169         lockdep_assert_held(&nn->client_lock);
 170
 171         if (!atomic_dec_and_test(&clp->cl_refcount))
 172                 return;
 173         if (!is_client_expired(clp))
 174                 renew_client_locked(clp);
 175 }
 176
 177 static void put_client_renew(struct nfs4_client *clp)
 178 {
 179         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 180
 181         if (!atomic_dec_and_lock(&clp->cl_refcount, &nn->client_lock))
 182                 return;
 183         if (!is_client_expired(clp))
 184                 renew_client_locked(clp);
 185         spin_unlock(&nn->client_lock);
 186 }
 187
 188 static __be32 nfsd4_get_session_locked(struct nfsd4_session *ses)
 189 {
 190         __be32 status;
 191
 192         if (is_session_dead(ses))
 193                 return nfserr_badsession;
 194         status = get_client_locked(ses->se_client);
 195         if (status)
 196                 return status;
 197         atomic_inc(&ses->se_ref);
 198         return nfs_ok;
 199 }
 200
 201 static void nfsd4_put_session_locked(struct nfsd4_session *ses)
 202 {
 203         struct nfs4_client *clp = ses->se_client;
 204         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 205
 206         lockdep_assert_held(&nn->client_lock);
 207
 208         if (atomic_dec_and_test(&ses->se_ref) && is_session_dead(ses))
 209                 free_session(ses);
 210         put_client_renew_locked(clp);
 211 }
 212
 213 static void nfsd4_put_session(struct nfsd4_session *ses)
 214 {
 215         struct nfs4_client *clp = ses->se_client;
 216         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
 217
 218         spin_lock(&nn->client_lock);
 219         nfsd4_put_session_locked(ses);
 220         spin_unlock(&nn->client_lock);
 221 }
 222
 223 static struct nfsd4_blocked_lock *
 224 find_blocked_lock(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 225                         struct nfsd_net *nn)
 226 {
 227         struct nfsd4_blocked_lock *cur, *found = NULL;
 228
 229         spin_lock(&nn->blocked_locks_lock);
 230         list_for_each_entry(cur, &lo->lo_blocked, nbl_list) {
 231                 if (fh_match(fh, &cur->nbl_fh)) {
 232                         list_del_init(&cur->nbl_list);
 233                         list_del_init(&cur->nbl_lru);
 234                         found = cur;
 235                         break;
 236                 }
 237         }
 238         spin_unlock(&nn->blocked_locks_lock);
 239         if (found)
 240                 posix_unblock_lock(&found->nbl_lock);
 241         return found;
 242 }
 243
 244 static struct nfsd4_blocked_lock *
 245 find_or_allocate_block(struct nfs4_lockowner *lo, struct knfsd_fh *fh,
 246                         struct nfsd_net *nn)
 247 {
 248         struct nfsd4_blocked_lock *nbl;
 249
 250         nbl = find_blocked_lock(lo, fh, nn);
 251         if (!nbl) {
 252                 nbl= kmalloc(sizeof(*nbl), GFP_KERNEL);
 253                 if (nbl) {
 254                         fh_copy_shallow(&nbl->nbl_fh, fh);
 255                         locks_init_lock(&nbl->nbl_lock);
 256                         nfsd4_init_cb(&nbl->nbl_cb, lo->lo_owner.so_client,
 257                                         &nfsd4_cb_notify_lock_ops,
 258                                         NFSPROC4_CLNT_CB_NOTIFY_LOCK);
 259                 }
 260         }
 261         return nbl;
 262 }
 263
 264 static void
 265 free_blocked_lock(struct nfsd4_blocked_lock *nbl)
 266 {
 267         locks_release_private(&nbl->nbl_lock);
 268         kfree(nbl);
 269 }
 270
 271 static int
 272 nfsd4_cb_notify_lock_done(struct nfsd4_callback *cb, struct rpc_task *task)
 273 {
 274         /*
 275          * Since this is just an optimization, we don't try very hard if it
 276          * turns out not to succeed. We'll requeue it on NFS4ERR_DELAY, and
 277          * just quit trying on anything else.
 278          */
 279         switch (task->tk_status) {
 280         case -NFS4ERR_DELAY:
 281                 rpc_delay(task, 1 * HZ);
 282                 return 0;
 283         default:
 284                 return 1;
 285         }
 286 }
 287
 288 static void
 289 nfsd4_cb_notify_lock_release(struct nfsd4_callback *cb)
 290 {
 291         struct nfsd4_blocked_lock       *nbl = container_of(cb,
 292                                                 struct nfsd4_blocked_lock, nbl_cb);
 293
 294         free_blocked_lock(nbl);
 295 }
 296
 297 static const struct nfsd4_callback_ops nfsd4_cb_notify_lock_ops = {
 298         .done           = nfsd4_cb_notify_lock_done,
 299         .release        = nfsd4_cb_notify_lock_release,
 300 };
 301
 302 static inline struct nfs4_stateowner *
 303 nfs4_get_stateowner(struct nfs4_stateowner *sop)
 304 {
 305         atomic_inc(&sop->so_count);
 306         return sop;
 307 }
 308
 309 static int
 310 same_owner_str(struct nfs4_stateowner *sop, struct xdr_netobj *owner)
 311 {
 312         return (sop->so_owner.len == owner->len) &&
 313                 0 == memcmp(sop->so_owner.data, owner->data, owner->len);
 314 }
 315
 316 static struct nfs4_openowner *
 317 find_openstateowner_str_locked(unsigned int hashval, struct nfsd4_open *open,
 318                         struct nfs4_client *clp)
 319 {
 320         struct nfs4_stateowner *so;
 321
 322         lockdep_assert_held(&clp->cl_lock);
 323
 324         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[hashval],
 325                             so_strhash) {
 326                 if (!so->so_is_open_owner)
 327                         continue;
 328                 if (same_owner_str(so, &open->op_owner))
 329                         return openowner(nfs4_get_stateowner(so));
 330         }
 331         return NULL;
 332 }
 333
 334 static struct nfs4_openowner *
 335 find_openstateowner_str(unsigned int hashval, struct nfsd4_open *open,
 336                         struct nfs4_client *clp)
 337 {
 338         struct nfs4_openowner *oo;
 339
 340         spin_lock(&clp->cl_lock);
 341         oo = find_openstateowner_str_locked(hashval, open, clp);
 342         spin_unlock(&clp->cl_lock);
 343         return oo;
 344 }
 345
 346 static inline u32
 347 opaque_hashval(const void *ptr, int nbytes)
 348 {
 349         unsigned char *cptr = (unsigned char *) ptr;
 350
 351         u32 x = 0;
 352         while (nbytes--) {
 353                 x *= 37;
 354                 x += *cptr++;
 355         }
 356         return x;
 357 }
 358
 359 static void nfsd4_free_file_rcu(struct rcu_head *rcu)
 360 {
 361         struct nfs4_file *fp = container_of(rcu, struct nfs4_file, fi_rcu);
 362
 363         kmem_cache_free(file_slab, fp);
 364 }
 365
 366 void
 367 put_nfs4_file(struct nfs4_file *fi)
 368 {
 369         might_lock(&state_lock);
 370
 371         if (refcount_dec_and_lock(&fi->fi_ref, &state_lock)) {
 372                 hlist_del_rcu(&fi->fi_hash);
 373                 spin_unlock(&state_lock);
 374                 WARN_ON_ONCE(!list_empty(&fi->fi_clnt_odstate));
 375                 WARN_ON_ONCE(!list_empty(&fi->fi_delegations));
 376                 call_rcu(&fi->fi_rcu, nfsd4_free_file_rcu);
 377         }
 378 }
 379
 380 static struct file *
 381 __nfs4_get_fd(struct nfs4_file *f, int oflag)
 382 {
 383         if (f->fi_fds[oflag])
 384                 return get_file(f->fi_fds[oflag]);
 385         return NULL;
 386 }
 387
 388 static struct file *
 389 find_writeable_file_locked(struct nfs4_file *f)
 390 {
 391         struct file *ret;
 392
 393         lockdep_assert_held(&f->fi_lock);
 394
 395         ret = __nfs4_get_fd(f, O_WRONLY);
 396         if (!ret)
 397                 ret = __nfs4_get_fd(f, O_RDWR);
 398         return ret;
 399 }
 400
 401 static struct file *
 402 find_writeable_file(struct nfs4_file *f)
 403 {
 404         struct file *ret;
 405
 406         spin_lock(&f->fi_lock);
 407         ret = find_writeable_file_locked(f);
 408         spin_unlock(&f->fi_lock);
 409
 410         return ret;
 411 }
 412
 413 static struct file *find_readable_file_locked(struct nfs4_file *f)
 414 {
 415         struct file *ret;
 416
 417         lockdep_assert_held(&f->fi_lock);
 418
 419         ret = __nfs4_get_fd(f, O_RDONLY);
 420         if (!ret)
 421                 ret = __nfs4_get_fd(f, O_RDWR);
 422         return ret;
 423 }
 424
 425 static struct file *
 426 find_readable_file(struct nfs4_file *f)
 427 {
 428         struct file *ret;
 429
 430         spin_lock(&f->fi_lock);
 431         ret = find_readable_file_locked(f);
 432         spin_unlock(&f->fi_lock);
 433
 434         return ret;
 435 }
 436
 437 struct file *
 438 find_any_file(struct nfs4_file *f)
 439 {
 440         struct file *ret;
 441
 442         spin_lock(&f->fi_lock);
 443         ret = __nfs4_get_fd(f, O_RDWR);
 444         if (!ret) {
 445                 ret = __nfs4_get_fd(f, O_WRONLY);
 446                 if (!ret)
 447                         ret = __nfs4_get_fd(f, O_RDONLY);
 448         }
 449         spin_unlock(&f->fi_lock);
 450         return ret;
 451 }
 452
 453 static atomic_long_t num_delegations;
 454 unsigned long max_delegations;
 455
 456 /*
 457  * Open owner state (share locks)
 458  */
 459
 460 /* hash tables for lock and open owners */
 461 #define OWNER_HASH_BITS              8
 462 #define OWNER_HASH_SIZE             (1 << OWNER_HASH_BITS)
 463 #define OWNER_HASH_MASK             (OWNER_HASH_SIZE - 1)
 464
 465 static unsigned int ownerstr_hashval(struct xdr_netobj *ownername)
 466 {
 467         unsigned int ret;
 468
 469         ret = opaque_hashval(ownername->data, ownername->len);
 470         return ret & OWNER_HASH_MASK;
 471 }
 472
 473 /* hash table for nfs4_file */
 474 #define FILE_HASH_BITS                   8
 475 #define FILE_HASH_SIZE                  (1 << FILE_HASH_BITS)
 476
 477 static unsigned int nfsd_fh_hashval(struct knfsd_fh *fh)
 478 {
 479         return jhash2(fh->fh_base.fh_pad, XDR_QUADLEN(fh->fh_size), 0);
 480 }
 481
 482 static unsigned int file_hashval(struct knfsd_fh *fh)
 483 {
 484         return nfsd_fh_hashval(fh) & (FILE_HASH_SIZE - 1);
 485 }
 486
 487 static struct hlist_head file_hashtbl[FILE_HASH_SIZE];
 488
 489 static void
 490 __nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 491 {
 492         lockdep_assert_held(&fp->fi_lock);
 493
 494         if (access & NFS4_SHARE_ACCESS_WRITE)
 495                 atomic_inc(&fp->fi_access[O_WRONLY]);
 496         if (access & NFS4_SHARE_ACCESS_READ)
 497                 atomic_inc(&fp->fi_access[O_RDONLY]);
 498 }
 499
 500 static __be32
 501 nfs4_file_get_access(struct nfs4_file *fp, u32 access)
 502 {
 503         lockdep_assert_held(&fp->fi_lock);
 504
 505         /* Does this access mode make sense? */
 506         if (access & ~NFS4_SHARE_ACCESS_BOTH)
 507                 return nfserr_inval;
 508
 509         /* Does it conflict with a deny mode already set? */
 510         if ((access & fp->fi_share_deny) != 0)
 511                 return nfserr_share_denied;
 512
 513         __nfs4_file_get_access(fp, access);
 514         return nfs_ok;
 515 }
 516
 517 static __be32 nfs4_file_check_deny(struct nfs4_file *fp, u32 deny)
 518 {
 519         /* Common case is that there is no deny mode. */
 520         if (deny) {
 521                 /* Does this deny mode make sense? */
 522                 if (deny & ~NFS4_SHARE_DENY_BOTH)
 523                         return nfserr_inval;
 524
 525                 if ((deny & NFS4_SHARE_DENY_READ) &&
 526                     atomic_read(&fp->fi_access[O_RDONLY]))
 527                         return nfserr_share_denied;
 528
 529                 if ((deny & NFS4_SHARE_DENY_WRITE) &&
 530                     atomic_read(&fp->fi_access[O_WRONLY]))
 531                         return nfserr_share_denied;
 532         }
 533         return nfs_ok;
 534 }
 535
 536 static void __nfs4_file_put_access(struct nfs4_file *fp, int oflag)
 537 {
 538         might_lock(&fp->fi_lock);
 539
 540         if (atomic_dec_and_lock(&fp->fi_access[oflag], &fp->fi_lock)) {
 541                 struct file *f1 = NULL;
 542                 struct file *f2 = NULL;
 543
 544                 swap(f1, fp->fi_fds[oflag]);
 545                 if (atomic_read(&fp->fi_access[1 - oflag]) == 0)
 546                         swap(f2, fp->fi_fds[O_RDWR]);
 547                 spin_unlock(&fp->fi_lock);
 548                 if (f1)
 549                         fput(f1);
 550                 if (f2)
 551                         fput(f2);
 552         }
 553 }
 554
 555 static void nfs4_file_put_access(struct nfs4_file *fp, u32 access)
 556 {
 557         WARN_ON_ONCE(access & ~NFS4_SHARE_ACCESS_BOTH);
 558
 559         if (access & NFS4_SHARE_ACCESS_WRITE)
 560                 __nfs4_file_put_access(fp, O_WRONLY);
 561         if (access & NFS4_SHARE_ACCESS_READ)
 562                 __nfs4_file_put_access(fp, O_RDONLY);
 563 }
 564
 565 /*
 566  * Allocate a new open/delegation state counter. This is needed for
 567  * pNFS for proper return on close semantics.
 568  *
 569  * Note that we only allocate it for pNFS-enabled exports, otherwise
 570  * all pointers to struct nfs4_clnt_odstate are always NULL.
 571  */
 572 static struct nfs4_clnt_odstate *
 573 alloc_clnt_odstate(struct nfs4_client *clp)
 574 {
 575         struct nfs4_clnt_odstate *co;
 576
 577         co = kmem_cache_zalloc(odstate_slab, GFP_KERNEL);
 578         if (co) {
 579                 co->co_client = clp;
 580                 refcount_set(&co->co_odcount, 1);
 581         }
 582         return co;
 583 }
 584
 585 static void
 586 hash_clnt_odstate_locked(struct nfs4_clnt_odstate *co)
 587 {
 588         struct nfs4_file *fp = co->co_file;
 589
 590         lockdep_assert_held(&fp->fi_lock);
 591         list_add(&co->co_perfile, &fp->fi_clnt_odstate);
 592 }
 593
 594 static inline void
 595 get_clnt_odstate(struct nfs4_clnt_odstate *co)
 596 {
 597         if (co)
 598                 refcount_inc(&co->co_odcount);
 599 }
 600
 601 static void
 602 put_clnt_odstate(struct nfs4_clnt_odstate *co)
 603 {
 604         struct nfs4_file *fp;
 605
 606         if (!co)
 607                 return;
 608
 609         fp = co->co_file;
 610         if (refcount_dec_and_lock(&co->co_odcount, &fp->fi_lock)) {
 611                 list_del(&co->co_perfile);
 612                 spin_unlock(&fp->fi_lock);
 613
 614                 nfsd4_return_all_file_layouts(co->co_client, fp);
 615                 kmem_cache_free(odstate_slab, co);
 616         }
 617 }
 618
 619 static struct nfs4_clnt_odstate *
 620 find_or_hash_clnt_odstate(struct nfs4_file *fp, struct nfs4_clnt_odstate *new)
 621 {
 622         struct nfs4_clnt_odstate *co;
 623         struct nfs4_client *cl;
 624
 625         if (!new)
 626                 return NULL;
 627
 628         cl = new->co_client;
 629
 630         spin_lock(&fp->fi_lock);
 631         list_for_each_entry(co, &fp->fi_clnt_odstate, co_perfile) {
 632                 if (co->co_client == cl) {
 633                         get_clnt_odstate(co);
 634                         goto out;
 635                 }
 636         }
 637         co = new;
 638         co->co_file = fp;
 639         hash_clnt_odstate_locked(new);
 640 out:
 641         spin_unlock(&fp->fi_lock);
 642         return co;
 643 }
 644
 645 struct nfs4_stid *nfs4_alloc_stid(struct nfs4_client *cl, struct kmem_cache *slab,
 646                                   void (*sc_free)(struct nfs4_stid *))
 647 {
 648         struct nfs4_stid *stid;
 649         int new_id;
 650
 651         stid = kmem_cache_zalloc(slab, GFP_KERNEL);
 652         if (!stid)
 653                 return NULL;
 654
 655         idr_preload(GFP_KERNEL);
 656         spin_lock(&cl->cl_lock);
 657         new_id = idr_alloc_cyclic(&cl->cl_stateids, stid, 0, 0, GFP_NOWAIT);
 658         spin_unlock(&cl->cl_lock);
 659         idr_preload_end();
 660         if (new_id < 0)
 661                 goto out_free;
 662
 663         stid->sc_free = sc_free;
 664         stid->sc_client = cl;
 665         stid->sc_stateid.si_opaque.so_id = new_id;
 666         stid->sc_stateid.si_opaque.so_clid = cl->cl_clientid;
 667         /* Will be incremented before return to client: */
 668         refcount_set(&stid->sc_count, 1);
 669         spin_lock_init(&stid->sc_lock);
 670
 671         /*
 672          * It shouldn't be a problem to reuse an opaque stateid value.
 673          * I don't think it is for 4.1.  But with 4.0 I worry that, for
 674          * example, a stray write retransmission could be accepted by
 675          * the server when it should have been rejected.  Therefore,
 676          * adopt a trick from the sctp code to attempt to maximize the
 677          * amount of time until an id is reused, by ensuring they always
 678          * "increase" (mod INT_MAX):
 679          */
 680         return stid;
 681 out_free:
 682         kmem_cache_free(slab, stid);
 683         return NULL;
 684 }
 685
 686 static struct nfs4_ol_stateid * nfs4_alloc_open_stateid(struct nfs4_client *clp)
 687 {
 688         struct nfs4_stid *stid;
 689
 690         stid = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_ol_stateid);
 691         if (!stid)
 692                 return NULL;
 693
 694         return openlockstateid(stid);
 695 }
 696
 697 static void nfs4_free_deleg(struct nfs4_stid *stid)
 698 {
 699         kmem_cache_free(deleg_slab, stid);
 700         atomic_long_dec(&num_delegations);
 701 }
 702
 703 /*
 704  * When we recall a delegation, we should be careful not to hand it
 705  * out again straight away.
 706  * To ensure this we keep a pair of bloom filters ('new' and 'old')
 707  * in which the filehandles of recalled delegations are "stored".
 708  * If a filehandle appear in either filter, a delegation is blocked.
 709  * When a delegation is recalled, the filehandle is stored in the "new"
 710  * filter.
 711  * Every 30 seconds we swap the filters and clear the "new" one,
 712  * unless both are empty of course.
 713  *
 714  * Each filter is 256 bits.  We hash the filehandle to 32bit and use the
 715  * low 3 bytes as hash-table indices.
 716  *
 717  * 'blocked_delegations_lock', which is always taken in block_delegations(),
 718  * is used to manage concurrent access.  Testing does not need the lock
 719  * except when swapping the two filters.
 720  */
 721 static DEFINE_SPINLOCK(blocked_delegations_lock);
 722 static struct bloom_pair {
 723         int     entries, old_entries;
 724         time_t  swap_time;
 725         int     new; /* index into 'set' */
 726         DECLARE_BITMAP(set[2], 256);
 727 } blocked_delegations;
 728
 729 static int delegation_blocked(struct knfsd_fh *fh)
 730 {
 731         u32 hash;
 732         struct bloom_pair *bd = &blocked_delegations;
 733
 734         if (bd->entries == 0)
 735                 return 0;
 736         if (seconds_since_boot() - bd->swap_time > 30) {
 737                 spin_lock(&blocked_delegations_lock);
 738                 if (seconds_since_boot() - bd->swap_time > 30) {
 739                         bd->entries -= bd->old_entries;
 740                         bd->old_entries = bd->entries;
 741                         memset(bd->set[bd->new], 0,
 742                                sizeof(bd->set[0]));
 743                         bd->new = 1-bd->new;
 744                         bd->swap_time = seconds_since_boot();
 745                 }
 746                 spin_unlock(&blocked_delegations_lock);
 747         }
 748         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 749         if (test_bit(hash&255, bd->set[0]) &&
 750             test_bit((hash>>8)&255, bd->set[0]) &&
 751             test_bit((hash>>16)&255, bd->set[0]))
 752                 return 1;
 753
 754         if (test_bit(hash&255, bd->set[1]) &&
 755             test_bit((hash>>8)&255, bd->set[1]) &&
 756             test_bit((hash>>16)&255, bd->set[1]))
 757                 return 1;
 758
 759         return 0;
 760 }
 761
 762 static void block_delegations(struct knfsd_fh *fh)
 763 {
 764         u32 hash;
 765         struct bloom_pair *bd = &blocked_delegations;
 766
 767         hash = jhash(&fh->fh_base, fh->fh_size, 0);
 768
 769         spin_lock(&blocked_delegations_lock);
 770         __set_bit(hash&255, bd->set[bd->new]);
 771         __set_bit((hash>>8)&255, bd->set[bd->new]);
 772         __set_bit((hash>>16)&255, bd->set[bd->new]);
 773         if (bd->entries == 0)
 774                 bd->swap_time = seconds_since_boot();
 775         bd->entries += 1;
 776         spin_unlock(&blocked_delegations_lock);
 777 }
 778
 779 static struct nfs4_delegation *
 780 alloc_init_deleg(struct nfs4_client *clp, struct svc_fh *current_fh,
 781                  struct nfs4_clnt_odstate *odstate)
 782 {
 783         struct nfs4_delegation *dp;
 784         long n;
 785
 786         dprintk("NFSD alloc_init_deleg\n");
 787         n = atomic_long_inc_return(&num_delegations);
 788         if (n < 0 || n > max_delegations)
 789                 goto out_dec;
 790         if (delegation_blocked(&current_fh->fh_handle))
 791                 goto out_dec;
 792         dp = delegstateid(nfs4_alloc_stid(clp, deleg_slab, nfs4_free_deleg));
 793         if (dp == NULL)
 794                 goto out_dec;
 795
 796         /*
 797          * delegation seqid's are never incremented.  The 4.1 special
 798          * meaning of seqid 0 isn't meaningful, really, but let's avoid
 799          * 0 anyway just for consistency and use 1:
 800          */
 801         dp->dl_stid.sc_stateid.si_generation = 1;
 802         INIT_LIST_HEAD(&dp->dl_perfile);
 803         INIT_LIST_HEAD(&dp->dl_perclnt);
 804         INIT_LIST_HEAD(&dp->dl_recall_lru);
 805         dp->dl_clnt_odstate = odstate;
 806         get_clnt_odstate(odstate);
 807         dp->dl_type = NFS4_OPEN_DELEGATE_READ;
 808         dp->dl_retries = 1;
 809         nfsd4_init_cb(&dp->dl_recall, dp->dl_stid.sc_client,
 810                       &nfsd4_cb_recall_ops, NFSPROC4_CLNT_CB_RECALL);
 811         return dp;
 812 out_dec:
 813         atomic_long_dec(&num_delegations);
 814         return NULL;
 815 }
 816
 817 void
 818 nfs4_put_stid(struct nfs4_stid *s)
 819 {
 820         struct nfs4_file *fp = s->sc_file;
 821         struct nfs4_client *clp = s->sc_client;
 822
 823         might_lock(&clp->cl_lock);
 824
 825         if (!refcount_dec_and_lock(&s->sc_count, &clp->cl_lock)) {
 826                 wake_up_all(&close_wq);
 827                 return;
 828         }
 829         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
 830         spin_unlock(&clp->cl_lock);
 831         s->sc_free(s);
 832         if (fp)
 833                 put_nfs4_file(fp);
 834 }
 835
 836 void
 837 nfs4_inc_and_copy_stateid(stateid_t *dst, struct nfs4_stid *stid)
 838 {
 839         stateid_t *src = &stid->sc_stateid;
 840
 841         spin_lock(&stid->sc_lock);
 842         if (unlikely(++src->si_generation == 0))
 843                 src->si_generation = 1;
 844         memcpy(dst, src, sizeof(*dst));
 845         spin_unlock(&stid->sc_lock);
 846 }
 847
 848 static void nfs4_put_deleg_lease(struct nfs4_file *fp)
 849 {
 850         struct file *filp = NULL;
 851
 852         spin_lock(&fp->fi_lock);
 853         if (fp->fi_deleg_file && --fp->fi_delegees == 0)
 854                 swap(filp, fp->fi_deleg_file);
 855         spin_unlock(&fp->fi_lock);
 856
 857         if (filp) {
 858                 vfs_setlease(filp, F_UNLCK, NULL, (void **)&fp);
 859                 fput(filp);
 860         }
 861 }
 862
 863 void nfs4_unhash_stid(struct nfs4_stid *s)
 864 {
 865         s->sc_type = 0;
 866 }
 867
 868 /**
 869  * nfs4_get_existing_delegation - Discover if this delegation already exists
 870  * @clp:     a pointer to the nfs4_client we're granting a delegation to
 871  * @fp:      a pointer to the nfs4_file we're granting a delegation on
 872  *
 873  * Return:
 874  *      On success: NULL if an existing delegation was not found.
 875  *
 876  *      On error: -EAGAIN if one was previously granted to this nfs4_client
 877  *                 for this nfs4_file.
 878  *
 879  */
 880
 881 static int
 882 nfs4_get_existing_delegation(struct nfs4_client *clp, struct nfs4_file *fp)
 883 {
 884         struct nfs4_delegation *searchdp = NULL;
 885         struct nfs4_client *searchclp = NULL;
 886
 887         lockdep_assert_held(&state_lock);
 888         lockdep_assert_held(&fp->fi_lock);
 889
 890         list_for_each_entry(searchdp, &fp->fi_delegations, dl_perfile) {
 891                 searchclp = searchdp->dl_stid.sc_client;
 892                 if (clp == searchclp) {
 893                         return -EAGAIN;
 894                 }
 895         }
 896         return 0;
 897 }
 898
 899 /**
 900  * hash_delegation_locked - Add a delegation to the appropriate lists
 901  * @dp:     a pointer to the nfs4_delegation we are adding.
 902  * @fp:     a pointer to the nfs4_file we're granting a delegation on
 903  *
 904  * Return:
 905  *      On success: NULL if the delegation was successfully hashed.
 906  *
 907  *      On error: -EAGAIN if one was previously granted to this
 908  *                 nfs4_client for this nfs4_file. Delegation is not hashed.
 909  *
 910  */
 911
 912 static int
 913 hash_delegation_locked(struct nfs4_delegation *dp, struct nfs4_file *fp)
 914 {
 915         int status;
 916         struct nfs4_client *clp = dp->dl_stid.sc_client;
 917
 918         lockdep_assert_held(&state_lock);
 919         lockdep_assert_held(&fp->fi_lock);
 920
 921         status = nfs4_get_existing_delegation(clp, fp);
 922         if (status)
 923                 return status;
 924         ++fp->fi_delegees;
 925         refcount_inc(&dp->dl_stid.sc_count);
 926         dp->dl_stid.sc_type = NFS4_DELEG_STID;
 927         list_add(&dp->dl_perfile, &fp->fi_delegations);
 928         list_add(&dp->dl_perclnt, &clp->cl_delegations);
 929         return 0;
 930 }
 931
 932 static bool
 933 unhash_delegation_locked(struct nfs4_delegation *dp)
 934 {
 935         struct nfs4_file *fp = dp->dl_stid.sc_file;
 936
 937         lockdep_assert_held(&state_lock);
 938
 939         if (list_empty(&dp->dl_perfile))
 940                 return false;
 941
 942         dp->dl_stid.sc_type = NFS4_CLOSED_DELEG_STID;
 943         /* Ensure that deleg break won't try to requeue it */
 944         ++dp->dl_time;
 945         spin_lock(&fp->fi_lock);
 946         list_del_init(&dp->dl_perclnt);
 947         list_del_init(&dp->dl_recall_lru);
 948         list_del_init(&dp->dl_perfile);
 949         spin_unlock(&fp->fi_lock);
 950         return true;
 951 }
 952
 953 static void destroy_delegation(struct nfs4_delegation *dp)
 954 {
 955         bool unhashed;
 956
 957         spin_lock(&state_lock);
 958         unhashed = unhash_delegation_locked(dp);
 959         spin_unlock(&state_lock);
 960         if (unhashed) {
 961                 put_clnt_odstate(dp->dl_clnt_odstate);
 962                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 963                 nfs4_put_stid(&dp->dl_stid);
 964         }
 965 }
 966
 967 static void revoke_delegation(struct nfs4_delegation *dp)
 968 {
 969         struct nfs4_client *clp = dp->dl_stid.sc_client;
 970
 971         WARN_ON(!list_empty(&dp->dl_recall_lru));
 972
 973         put_clnt_odstate(dp->dl_clnt_odstate);
 974         nfs4_put_deleg_lease(dp->dl_stid.sc_file);
 975
 976         if (clp->cl_minorversion == 0)
 977                 nfs4_put_stid(&dp->dl_stid);
 978         else {
 979                 dp->dl_stid.sc_type = NFS4_REVOKED_DELEG_STID;
 980                 spin_lock(&clp->cl_lock);
 981                 list_add(&dp->dl_recall_lru, &clp->cl_revoked);
 982                 spin_unlock(&clp->cl_lock);
 983         }
 984 }
 985
 986 /*
 987  * SETCLIENTID state
 988  */
 989
 990 static unsigned int clientid_hashval(u32 id)
 991 {
 992         return id & CLIENT_HASH_MASK;
 993 }
 994
 995 static unsigned int clientstr_hashval(const char *name)
 996 {
 997         return opaque_hashval(name, 8) & CLIENT_HASH_MASK;
 998 }
 999
1000 /*
1001  * We store the NONE, READ, WRITE, and BOTH bits separately in the
1002  * st_{access,deny}_bmap field of the stateid, in order to track not
1003  * only what share bits are currently in force, but also what
1004  * combinations of share bits previous opens have used.  This allows us
1005  * to enforce the recommendation of rfc 3530 14.2.19 that the server
1006  * return an error if the client attempt to downgrade to a combination
1007  * of share bits not explicable by closing some of its previous opens.
1008  *
1009  * XXX: This enforcement is actually incomplete, since we don't keep
1010  * track of access/deny bit combinations; so, e.g., we allow:
1011  *
1012  *      OPEN allow read, deny write
1013  *      OPEN allow both, deny none
1014  *      DOWNGRADE allow read, deny none
1015  *
1016  * which we should reject.
1017  */
1018 static unsigned int
1019 bmap_to_share_mode(unsigned long bmap) {
1020         int i;
1021         unsigned int access = 0;
1022
1023         for (i = 1; i < 4; i++) {
1024                 if (test_bit(i, &bmap))
1025                         access |= i;
1026         }
1027         return access;
1028 }
1029
1030 /* set share access for a given stateid */
1031 static inline void
1032 set_access(u32 access, struct nfs4_ol_stateid *stp)
1033 {
1034         unsigned char mask = 1 << access;
1035
1036         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1037         stp->st_access_bmap |= mask;
1038 }
1039
1040 /* clear share access for a given stateid */
1041 static inline void
1042 clear_access(u32 access, struct nfs4_ol_stateid *stp)
1043 {
1044         unsigned char mask = 1 << access;
1045
1046         WARN_ON_ONCE(access > NFS4_SHARE_ACCESS_BOTH);
1047         stp->st_access_bmap &= ~mask;
1048 }
1049
1050 /* test whether a given stateid has access */
1051 static inline bool
1052 test_access(u32 access, struct nfs4_ol_stateid *stp)
1053 {
1054         unsigned char mask = 1 << access;
1055
1056         return (bool)(stp->st_access_bmap & mask);
1057 }
1058
1059 /* set share deny for a given stateid */
1060 static inline void
1061 set_deny(u32 deny, struct nfs4_ol_stateid *stp)
1062 {
1063         unsigned char mask = 1 << deny;
1064
1065         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1066         stp->st_deny_bmap |= mask;
1067 }
1068
1069 /* clear share deny for a given stateid */
1070 static inline void
1071 clear_deny(u32 deny, struct nfs4_ol_stateid *stp)
1072 {
1073         unsigned char mask = 1 << deny;
1074
1075         WARN_ON_ONCE(deny > NFS4_SHARE_DENY_BOTH);
1076         stp->st_deny_bmap &= ~mask;
1077 }
1078
1079 /* test whether a given stateid is denying specific access */
1080 static inline bool
1081 test_deny(u32 deny, struct nfs4_ol_stateid *stp)
1082 {
1083         unsigned char mask = 1 << deny;
1084
1085         return (bool)(stp->st_deny_bmap & mask);
1086 }
1087
1088 static int nfs4_access_to_omode(u32 access)
1089 {
1090         switch (access & NFS4_SHARE_ACCESS_BOTH) {
1091         case NFS4_SHARE_ACCESS_READ:
1092                 return O_RDONLY;
1093         case NFS4_SHARE_ACCESS_WRITE:
1094                 return O_WRONLY;
1095         case NFS4_SHARE_ACCESS_BOTH:
1096                 return O_RDWR;
1097         }
1098         WARN_ON_ONCE(1);
1099         return O_RDONLY;
1100 }
1101
1102 /*
1103  * A stateid that had a deny mode associated with it is being released
1104  * or downgraded. Recalculate the deny mode on the file.
1105  */
1106 static void
1107 recalculate_deny_mode(struct nfs4_file *fp)
1108 {
1109         struct nfs4_ol_stateid *stp;
1110
1111         spin_lock(&fp->fi_lock);
1112         fp->fi_share_deny = 0;
1113         list_for_each_entry(stp, &fp->fi_stateids, st_perfile)
1114                 fp->fi_share_deny |= bmap_to_share_mode(stp->st_deny_bmap);
1115         spin_unlock(&fp->fi_lock);
1116 }
1117
1118 static void
1119 reset_union_bmap_deny(u32 deny, struct nfs4_ol_stateid *stp)
1120 {
1121         int i;
1122         bool change = false;
1123
1124         for (i = 1; i < 4; i++) {
1125                 if ((i & deny) != i) {
1126                         change = true;
1127                         clear_deny(i, stp);
1128                 }
1129         }
1130
1131         /* Recalculate per-file deny mode if there was a change */
1132         if (change)
1133                 recalculate_deny_mode(stp->st_stid.sc_file);
1134 }
1135
1136 /* release all access and file references for a given stateid */
1137 static void
1138 release_all_access(struct nfs4_ol_stateid *stp)
1139 {
1140         int i;
1141         struct nfs4_file *fp = stp->st_stid.sc_file;
1142
1143         if (fp && stp->st_deny_bmap != 0)
1144                 recalculate_deny_mode(fp);
1145
1146         for (i = 1; i < 4; i++) {
1147                 if (test_access(i, stp))
1148                         nfs4_file_put_access(stp->st_stid.sc_file, i);
1149                 clear_access(i, stp);
1150         }
1151 }
1152
1153 static inline void nfs4_free_stateowner(struct nfs4_stateowner *sop)
1154 {
1155         kfree(sop->so_owner.data);
1156         sop->so_ops->so_free(sop);
1157 }
1158
1159 static void nfs4_put_stateowner(struct nfs4_stateowner *sop)
1160 {
1161         struct nfs4_client *clp = sop->so_client;
1162
1163         might_lock(&clp->cl_lock);
1164
1165         if (!atomic_dec_and_lock(&sop->so_count, &clp->cl_lock))
1166                 return;
1167         sop->so_ops->so_unhash(sop);
1168         spin_unlock(&clp->cl_lock);
1169         nfs4_free_stateowner(sop);
1170 }
1171
1172 static bool unhash_ol_stateid(struct nfs4_ol_stateid *stp)
1173 {
1174         struct nfs4_file *fp = stp->st_stid.sc_file;
1175
1176         lockdep_assert_held(&stp->st_stateowner->so_client->cl_lock);
1177
1178         if (list_empty(&stp->st_perfile))
1179                 return false;
1180
1181         spin_lock(&fp->fi_lock);
1182         list_del_init(&stp->st_perfile);
1183         spin_unlock(&fp->fi_lock);
1184         list_del(&stp->st_perstateowner);
1185         return true;
1186 }
1187
1188 static void nfs4_free_ol_stateid(struct nfs4_stid *stid)
1189 {
1190         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1191
1192         put_clnt_odstate(stp->st_clnt_odstate);
1193         release_all_access(stp);
1194         if (stp->st_stateowner)
1195                 nfs4_put_stateowner(stp->st_stateowner);
1196         kmem_cache_free(stateid_slab, stid);
1197 }
1198
1199 static void nfs4_free_lock_stateid(struct nfs4_stid *stid)
1200 {
1201         struct nfs4_ol_stateid *stp = openlockstateid(stid);
1202         struct nfs4_lockowner *lo = lockowner(stp->st_stateowner);
1203         struct file *file;
1204
1205         file = find_any_file(stp->st_stid.sc_file);
1206         if (file)
1207                 filp_close(file, (fl_owner_t)lo);
1208         nfs4_free_ol_stateid(stid);
1209 }
1210
1211 /*
1212  * Put the persistent reference to an already unhashed generic stateid, while
1213  * holding the cl_lock. If it's the last reference, then put it onto the
1214  * reaplist for later destruction.
1215  */
1216 static void put_ol_stateid_locked(struct nfs4_ol_stateid *stp,
1217                                        struct list_head *reaplist)
1218 {
1219         struct nfs4_stid *s = &stp->st_stid;
1220         struct nfs4_client *clp = s->sc_client;
1221
1222         lockdep_assert_held(&clp->cl_lock);
1223
1224         WARN_ON_ONCE(!list_empty(&stp->st_locks));
1225
1226         if (!refcount_dec_and_test(&s->sc_count)) {
1227                 wake_up_all(&close_wq);
1228                 return;
1229         }
1230
1231         idr_remove(&clp->cl_stateids, s->sc_stateid.si_opaque.so_id);
1232         list_add(&stp->st_locks, reaplist);
1233 }
1234
1235 static bool unhash_lock_stateid(struct nfs4_ol_stateid *stp)
1236 {
1237         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1238
1239         list_del_init(&stp->st_locks);
1240         nfs4_unhash_stid(&stp->st_stid);
1241         return unhash_ol_stateid(stp);
1242 }
1243
1244 static void release_lock_stateid(struct nfs4_ol_stateid *stp)
1245 {
1246         struct nfs4_client *clp = stp->st_stid.sc_client;
1247         bool unhashed;
1248
1249         spin_lock(&clp->cl_lock);
1250         unhashed = unhash_lock_stateid(stp);
1251         spin_unlock(&clp->cl_lock);
1252         if (unhashed)
1253                 nfs4_put_stid(&stp->st_stid);
1254 }
1255
1256 static void unhash_lockowner_locked(struct nfs4_lockowner *lo)
1257 {
1258         struct nfs4_client *clp = lo->lo_owner.so_client;
1259
1260         lockdep_assert_held(&clp->cl_lock);
1261
1262         list_del_init(&lo->lo_owner.so_strhash);
1263 }
1264
1265 /*
1266  * Free a list of generic stateids that were collected earlier after being
1267  * fully unhashed.
1268  */
1269 static void
1270 free_ol_stateid_reaplist(struct list_head *reaplist)
1271 {
1272         struct nfs4_ol_stateid *stp;
1273         struct nfs4_file *fp;
1274
1275         might_sleep();
1276
1277         while (!list_empty(reaplist)) {
1278                 stp = list_first_entry(reaplist, struct nfs4_ol_stateid,
1279                                        st_locks);
1280                 list_del(&stp->st_locks);
1281                 fp = stp->st_stid.sc_file;
1282                 stp->st_stid.sc_free(&stp->st_stid);
1283                 if (fp)
1284                         put_nfs4_file(fp);
1285         }
1286 }
1287
1288 static void release_open_stateid_locks(struct nfs4_ol_stateid *open_stp,
1289                                        struct list_head *reaplist)
1290 {
1291         struct nfs4_ol_stateid *stp;
1292
1293         lockdep_assert_held(&open_stp->st_stid.sc_client->cl_lock);
1294
1295         while (!list_empty(&open_stp->st_locks)) {
1296                 stp = list_entry(open_stp->st_locks.next,
1297                                 struct nfs4_ol_stateid, st_locks);
1298                 WARN_ON(!unhash_lock_stateid(stp));
1299                 put_ol_stateid_locked(stp, reaplist);
1300         }
1301 }
1302
1303 static bool unhash_open_stateid(struct nfs4_ol_stateid *stp,
1304                                 struct list_head *reaplist)
1305 {
1306         bool unhashed;
1307
1308         lockdep_assert_held(&stp->st_stid.sc_client->cl_lock);
1309
1310         unhashed = unhash_ol_stateid(stp);
1311         release_open_stateid_locks(stp, reaplist);
1312         return unhashed;
1313 }
1314
1315 static void release_open_stateid(struct nfs4_ol_stateid *stp)
1316 {
1317         LIST_HEAD(reaplist);
1318
1319         spin_lock(&stp->st_stid.sc_client->cl_lock);
1320         if (unhash_open_stateid(stp, &reaplist))
1321                 put_ol_stateid_locked(stp, &reaplist);
1322         spin_unlock(&stp->st_stid.sc_client->cl_lock);
1323         free_ol_stateid_reaplist(&reaplist);
1324 }
1325
1326 static void unhash_openowner_locked(struct nfs4_openowner *oo)
1327 {
1328         struct nfs4_client *clp = oo->oo_owner.so_client;
1329
1330         lockdep_assert_held(&clp->cl_lock);
1331
1332         list_del_init(&oo->oo_owner.so_strhash);
1333         list_del_init(&oo->oo_perclient);
1334 }
1335
1336 static void release_last_closed_stateid(struct nfs4_openowner *oo)
1337 {
1338         struct nfsd_net *nn = net_generic(oo->oo_owner.so_client->net,
1339                                           nfsd_net_id);
1340         struct nfs4_ol_stateid *s;
1341
1342         spin_lock(&nn->client_lock);
1343         s = oo->oo_last_closed_stid;
1344         if (s) {
1345                 list_del_init(&oo->oo_close_lru);
1346                 oo->oo_last_closed_stid = NULL;
1347         }
1348         spin_unlock(&nn->client_lock);
1349         if (s)
1350                 nfs4_put_stid(&s->st_stid);
1351 }
1352
1353 static void release_openowner(struct nfs4_openowner *oo)
1354 {
1355         struct nfs4_ol_stateid *stp;
1356         struct nfs4_client *clp = oo->oo_owner.so_client;
1357         struct list_head reaplist;
1358
1359         INIT_LIST_HEAD(&reaplist);
1360
1361         spin_lock(&clp->cl_lock);
1362         unhash_openowner_locked(oo);
1363         while (!list_empty(&oo->oo_owner.so_stateids)) {
1364                 stp = list_first_entry(&oo->oo_owner.so_stateids,
1365                                 struct nfs4_ol_stateid, st_perstateowner);
1366                 if (unhash_open_stateid(stp, &reaplist))
1367                         put_ol_stateid_locked(stp, &reaplist);
1368         }
1369         spin_unlock(&clp->cl_lock);
1370         free_ol_stateid_reaplist(&reaplist);
1371         release_last_closed_stateid(oo);
1372         nfs4_put_stateowner(&oo->oo_owner);
1373 }
1374
1375 static inline int
1376 hash_sessionid(struct nfs4_sessionid *sessionid)
1377 {
1378         struct nfsd4_sessionid *sid = (struct nfsd4_sessionid *)sessionid;
1379
1380         return sid->sequence % SESSION_HASH_SIZE;
1381 }
1382
1383 #ifdef CONFIG_SUNRPC_DEBUG
1384 static inline void
1385 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1386 {
1387         u32 *ptr = (u32 *)(&sessionid->data[0]);
1388         dprintk("%s: %u:%u:%u:%u\n", fn, ptr[0], ptr[1], ptr[2], ptr[3]);
1389 }
1390 #else
1391 static inline void
1392 dump_sessionid(const char *fn, struct nfs4_sessionid *sessionid)
1393 {
1394 }
1395 #endif
1396
1397 /*
1398  * Bump the seqid on cstate->replay_owner, and clear replay_owner if it
1399  * won't be used for replay.
1400  */
1401 void nfsd4_bump_seqid(struct nfsd4_compound_state *cstate, __be32 nfserr)
1402 {
1403         struct nfs4_stateowner *so = cstate->replay_owner;
1404
1405         if (nfserr == nfserr_replay_me)
1406                 return;
1407
1408         if (!seqid_mutating_err(ntohl(nfserr))) {
1409                 nfsd4_cstate_clear_replay(cstate);
1410                 return;
1411         }
1412         if (!so)
1413                 return;
1414         if (so->so_is_open_owner)
1415                 release_last_closed_stateid(openowner(so));
1416         so->so_seqid++;
1417         return;
1418 }
1419
1420 static void
1421 gen_sessionid(struct nfsd4_session *ses)
1422 {
1423         struct nfs4_client *clp = ses->se_client;
1424         struct nfsd4_sessionid *sid;
1425
1426         sid = (struct nfsd4_sessionid *)ses->se_sessionid.data;
1427         sid->clientid = clp->cl_clientid;
1428         sid->sequence = current_sessionid++;
1429         sid->reserved = 0;
1430 }
1431
1432 /*
1433  * The protocol defines ca_maxresponssize_cached to include the size of
1434  * the rpc header, but all we need to cache is the data starting after
1435  * the end of the initial SEQUENCE operation--the rest we regenerate
1436  * each time.  Therefore we can advertise a ca_maxresponssize_cached
1437  * value that is the number of bytes in our cache plus a few additional
1438  * bytes.  In order to stay on the safe side, and not promise more than
1439  * we can cache, those additional bytes must be the minimum possible: 24
1440  * bytes of rpc header (xid through accept state, with AUTH_NULL
1441  * verifier), 12 for the compound header (with zero-length tag), and 44
1442  * for the SEQUENCE op response:
1443  */
1444 #define NFSD_MIN_HDR_SEQ_SZ  (24 + 12 + 44)
1445
1446 static void
1447 free_session_slots(struct nfsd4_session *ses)
1448 {
1449         int i;
1450
1451         for (i = 0; i < ses->se_fchannel.maxreqs; i++) {
1452                 free_svc_cred(&ses->se_slots[i]->sl_cred);
1453                 kfree(ses->se_slots[i]);
1454         }
1455 }
1456
1457 /*
1458  * We don't actually need to cache the rpc and session headers, so we
1459  * can allocate a little less for each slot:
1460  */
1461 static inline u32 slot_bytes(struct nfsd4_channel_attrs *ca)
1462 {
1463         u32 size;
1464
1465         if (ca->maxresp_cached < NFSD_MIN_HDR_SEQ_SZ)
1466                 size = 0;
1467         else
1468                 size = ca->maxresp_cached - NFSD_MIN_HDR_SEQ_SZ;
1469         return size + sizeof(struct nfsd4_slot);
1470 }
1471
1472 /*
1473  * XXX: If we run out of reserved DRC memory we could (up to a point)
1474  * re-negotiate active sessions and reduce their slot usage to make
1475  * room for new connections. For now we just fail the create session.
1476  */
1477 static u32 nfsd4_get_drc_mem(struct nfsd4_channel_attrs *ca)
1478 {
1479         u32 slotsize = slot_bytes(ca);
1480         u32 num = ca->maxreqs;
1481         int avail;
1482
1483         spin_lock(&nfsd_drc_lock);
1484         avail = min((unsigned long)NFSD_MAX_MEM_PER_SESSION,
1485                     nfsd_drc_max_mem - nfsd_drc_mem_used);
1486         /*
1487          * Never use more than a third of the remaining memory,
1488          * unless it's the only way to give this client a slot:
1489          */
1490         avail = clamp_t(int, avail, slotsize, avail/3);
1491         num = min_t(int, num, avail / slotsize);
1492         nfsd_drc_mem_used += num * slotsize;
1493         spin_unlock(&nfsd_drc_lock);
1494
1495         return num;
1496 }
1497
1498 static void nfsd4_put_drc_mem(struct nfsd4_channel_attrs *ca)
1499 {
1500         int slotsize = slot_bytes(ca);
1501
1502         spin_lock(&nfsd_drc_lock);
1503         nfsd_drc_mem_used -= slotsize * ca->maxreqs;
1504         spin_unlock(&nfsd_drc_lock);
1505 }
1506
1507 static struct nfsd4_session *alloc_session(struct nfsd4_channel_attrs *fattrs,
1508                                            struct nfsd4_channel_attrs *battrs)
1509 {
1510         int numslots = fattrs->maxreqs;
1511         int slotsize = slot_bytes(fattrs);
1512         struct nfsd4_session *new;
1513         int mem, i;
1514
1515         BUILD_BUG_ON(NFSD_MAX_SLOTS_PER_SESSION * sizeof(struct nfsd4_slot *)
1516                         + sizeof(struct nfsd4_session) > PAGE_SIZE);
1517         mem = numslots * sizeof(struct nfsd4_slot *);
1518
1519         new = kzalloc(sizeof(*new) + mem, GFP_KERNEL);
1520         if (!new)
1521                 return NULL;
1522         /* allocate each struct nfsd4_slot and data cache in one piece */
1523         for (i = 0; i < numslots; i++) {
1524                 new->se_slots[i] = kzalloc(slotsize, GFP_KERNEL);
1525                 if (!new->se_slots[i])
1526                         goto out_free;
1527         }
1528
1529         memcpy(&new->se_fchannel, fattrs, sizeof(struct nfsd4_channel_attrs));
1530         memcpy(&new->se_bchannel, battrs, sizeof(struct nfsd4_channel_attrs));
1531
1532         return new;
1533 out_free:
1534         while (i--)
1535                 kfree(new->se_slots[i]);
1536         kfree(new);
1537         return NULL;
1538 }
1539
1540 static void free_conn(struct nfsd4_conn *c)
1541 {
1542         svc_xprt_put(c->cn_xprt);
1543         kfree(c);
1544 }
1545
1546 static void nfsd4_conn_lost(struct svc_xpt_user *u)
1547 {
1548         struct nfsd4_conn *c = container_of(u, struct nfsd4_conn, cn_xpt_user);
1549         struct nfs4_client *clp = c->cn_session->se_client;
1550
1551         spin_lock(&clp->cl_lock);
1552         if (!list_empty(&c->cn_persession)) {
1553                 list_del(&c->cn_persession);
1554                 free_conn(c);
1555         }
1556         nfsd4_probe_callback(clp);
1557         spin_unlock(&clp->cl_lock);
1558 }
1559
1560 static struct nfsd4_conn *alloc_conn(struct svc_rqst *rqstp, u32 flags)
1561 {
1562         struct nfsd4_conn *conn;
1563
1564         conn = kmalloc(sizeof(struct nfsd4_conn), GFP_KERNEL);
1565         if (!conn)
1566                 return NULL;
1567         svc_xprt_get(rqstp->rq_xprt);
1568         conn->cn_xprt = rqstp->rq_xprt;
1569         conn->cn_flags = flags;
1570         INIT_LIST_HEAD(&conn->cn_xpt_user.list);
1571         return conn;
1572 }
1573
1574 static void __nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1575 {
1576         conn->cn_session = ses;
1577         list_add(&conn->cn_persession, &ses->se_conns);
1578 }
1579
1580 static void nfsd4_hash_conn(struct nfsd4_conn *conn, struct nfsd4_session *ses)
1581 {
1582         struct nfs4_client *clp = ses->se_client;
1583
1584         spin_lock(&clp->cl_lock);
1585         __nfsd4_hash_conn(conn, ses);
1586         spin_unlock(&clp->cl_lock);
1587 }
1588
1589 static int nfsd4_register_conn(struct nfsd4_conn *conn)
1590 {
1591         conn->cn_xpt_user.callback = nfsd4_conn_lost;
1592         return register_xpt_user(conn->cn_xprt, &conn->cn_xpt_user);
1593 }
1594
1595 static void nfsd4_init_conn(struct svc_rqst *rqstp, struct nfsd4_conn *conn, struct nfsd4_session *ses)
1596 {
1597         int ret;
1598
1599         nfsd4_hash_conn(conn, ses);
1600         ret = nfsd4_register_conn(conn);
1601         if (ret)
1602                 /* oops; xprt is already down: */
1603                 nfsd4_conn_lost(&conn->cn_xpt_user);
1604         /* We may have gained or lost a callback channel: */
1605         nfsd4_probe_callback_sync(ses->se_client);
1606 }
1607
1608 static struct nfsd4_conn *alloc_conn_from_crses(struct svc_rqst *rqstp, struct nfsd4_create_session *cses)
1609 {
1610         u32 dir = NFS4_CDFC4_FORE;
1611
1612         if (cses->flags & SESSION4_BACK_CHAN)
1613                 dir |= NFS4_CDFC4_BACK;
1614         return alloc_conn(rqstp, dir);
1615 }
1616
1617 /* must be called under client_lock */
1618 static void nfsd4_del_conns(struct nfsd4_session *s)
1619 {
1620         struct nfs4_client *clp = s->se_client;
1621         struct nfsd4_conn *c;
1622
1623         spin_lock(&clp->cl_lock);
1624         while (!list_empty(&s->se_conns)) {
1625                 c = list_first_entry(&s->se_conns, struct nfsd4_conn, cn_persession);
1626                 list_del_init(&c->cn_persession);
1627                 spin_unlock(&clp->cl_lock);
1628
1629                 unregister_xpt_user(c->cn_xprt, &c->cn_xpt_user);
1630                 free_conn(c);
1631
1632                 spin_lock(&clp->cl_lock);
1633         }
1634         spin_unlock(&clp->cl_lock);
1635 }
1636
1637 static void __free_session(struct nfsd4_session *ses)
1638 {
1639         free_session_slots(ses);
1640         kfree(ses);
1641 }
1642
1643 static void free_session(struct nfsd4_session *ses)
1644 {
1645         nfsd4_del_conns(ses);
1646         nfsd4_put_drc_mem(&ses->se_fchannel);
1647         __free_session(ses);
1648 }
1649
1650 static void init_session(struct svc_rqst *rqstp, struct nfsd4_session *new, struct nfs4_client *clp, struct nfsd4_create_session *cses)
1651 {
1652         int idx;
1653         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
1654
1655         new->se_client = clp;
1656         gen_sessionid(new);
1657
1658         INIT_LIST_HEAD(&new->se_conns);
1659
1660         new->se_cb_seq_nr = 1;
1661         new->se_flags = cses->flags;
1662         new->se_cb_prog = cses->callback_prog;
1663         new->se_cb_sec = cses->cb_sec;
1664         atomic_set(&new->se_ref, 0);
1665         idx = hash_sessionid(&new->se_sessionid);
1666         list_add(&new->se_hash, &nn->sessionid_hashtbl[idx]);
1667         spin_lock(&clp->cl_lock);
1668         list_add(&new->se_perclnt, &clp->cl_sessions);
1669         spin_unlock(&clp->cl_lock);
1670
1671         {
1672                 struct sockaddr *sa = svc_addr(rqstp);
1673                 /*
1674                  * This is a little silly; with sessions there's no real
1675                  * use for the callback address.  Use the peer address
1676                  * as a reasonable default for now, but consider fixing
1677                  * the rpc client not to require an address in the
1678                  * future:
1679                  */
1680                 rpc_copy_addr((struct sockaddr *)&clp->cl_cb_conn.cb_addr, sa);
1681                 clp->cl_cb_conn.cb_addrlen = svc_addr_len(sa);
1682         }
1683 }
1684
1685 /* caller must hold client_lock */
1686 static struct nfsd4_session *
1687 __find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net)
1688 {
1689         struct nfsd4_session *elem;
1690         int idx;
1691         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
1692
1693         lockdep_assert_held(&nn->client_lock);
1694
1695         dump_sessionid(__func__, sessionid);
1696         idx = hash_sessionid(sessionid);
1697         /* Search in the appropriate list */
1698         list_for_each_entry(elem, &nn->sessionid_hashtbl[idx], se_hash) {
1699                 if (!memcmp(elem->se_sessionid.data, sessionid->data,
1700                             NFS4_MAX_SESSIONID_LEN)) {
1701                         return elem;
1702                 }
1703         }
1704
1705         dprintk("%s: session not found\n", __func__);
1706         return NULL;
1707 }
1708
1709 static struct nfsd4_session *
1710 find_in_sessionid_hashtbl(struct nfs4_sessionid *sessionid, struct net *net,
1711                 __be32 *ret)
1712 {
1713         struct nfsd4_session *session;
1714         __be32 status = nfserr_badsession;
1715
1716         session = __find_in_sessionid_hashtbl(sessionid, net);
1717         if (!session)
1718                 goto out;
1719         status = nfsd4_get_session_locked(session);
1720         if (status)
1721                 session = NULL;
1722 out:
1723         *ret = status;
1724         return session;
1725 }
1726
1727 /* caller must hold client_lock */
1728 static void
1729 unhash_session(struct nfsd4_session *ses)
1730 {
1731         struct nfs4_client *clp = ses->se_client;
1732         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1733
1734         lockdep_assert_held(&nn->client_lock);
1735
1736         list_del(&ses->se_hash);
1737         spin_lock(&ses->se_client->cl_lock);
1738         list_del(&ses->se_perclnt);
1739         spin_unlock(&ses->se_client->cl_lock);
1740 }
1741
1742 /* SETCLIENTID and SETCLIENTID_CONFIRM Helper functions */
1743 static int
1744 STALE_CLIENTID(clientid_t *clid, struct nfsd_net *nn)
1745 {
1746         /*
1747          * We're assuming the clid was not given out from a boot
1748          * precisely 2^32 (about 136 years) before this one.  That seems
1749          * a safe assumption:
1750          */
1751         if (clid->cl_boot == (u32)nn->boot_time)
1752                 return 0;
1753         dprintk("NFSD stale clientid (%08x/%08x) boot_time %08lx\n",
1754                 clid->cl_boot, clid->cl_id, nn->boot_time);
1755         return 1;
1756 }
1757
1758 /*
1759  * XXX Should we use a slab cache ?
1760  * This type of memory management is somewhat inefficient, but we use it
1761  * anyway since SETCLIENTID is not a common operation.
1762  */
1763 static struct nfs4_client *alloc_client(struct xdr_netobj name)
1764 {
1765         struct nfs4_client *clp;
1766         int i;
1767
1768         clp = kzalloc(sizeof(struct nfs4_client), GFP_KERNEL);
1769         if (clp == NULL)
1770                 return NULL;
1771         clp->cl_name.data = kmemdup(name.data, name.len, GFP_KERNEL);
1772         if (clp->cl_name.data == NULL)
1773                 goto err_no_name;
1774         clp->cl_ownerstr_hashtbl = kmalloc(sizeof(struct list_head) *
1775                         OWNER_HASH_SIZE, GFP_KERNEL);
1776         if (!clp->cl_ownerstr_hashtbl)
1777                 goto err_no_hashtbl;
1778         for (i = 0; i < OWNER_HASH_SIZE; i++)
1779                 INIT_LIST_HEAD(&clp->cl_ownerstr_hashtbl[i]);
1780         clp->cl_name.len = name.len;
1781         INIT_LIST_HEAD(&clp->cl_sessions);
1782         idr_init(&clp->cl_stateids);
1783         atomic_set(&clp->cl_refcount, 0);
1784         clp->cl_cb_state = NFSD4_CB_UNKNOWN;
1785         INIT_LIST_HEAD(&clp->cl_idhash);
1786         INIT_LIST_HEAD(&clp->cl_openowners);
1787         INIT_LIST_HEAD(&clp->cl_delegations);
1788         INIT_LIST_HEAD(&clp->cl_lru);
1789         INIT_LIST_HEAD(&clp->cl_revoked);
1790 #ifdef CONFIG_NFSD_PNFS
1791         INIT_LIST_HEAD(&clp->cl_lo_states);
1792 #endif
1793         spin_lock_init(&clp->cl_lock);
1794         rpc_init_wait_queue(&clp->cl_cb_waitq, "Backchannel slot table");
1795         return clp;
1796 err_no_hashtbl:
1797         kfree(clp->cl_name.data);
1798 err_no_name:
1799         kfree(clp);
1800         return NULL;
1801 }
1802
1803 static void
1804 free_client(struct nfs4_client *clp)
1805 {
1806         while (!list_empty(&clp->cl_sessions)) {
1807                 struct nfsd4_session *ses;
1808                 ses = list_entry(clp->cl_sessions.next, struct nfsd4_session,
1809                                 se_perclnt);
1810                 list_del(&ses->se_perclnt);
1811                 WARN_ON_ONCE(atomic_read(&ses->se_ref));
1812                 free_session(ses);
1813         }
1814         rpc_destroy_wait_queue(&clp->cl_cb_waitq);
1815         free_svc_cred(&clp->cl_cred);
1816         kfree(clp->cl_ownerstr_hashtbl);
1817         kfree(clp->cl_name.data);
1818         idr_destroy(&clp->cl_stateids);
1819         kfree(clp);
1820 }
1821
1822 /* must be called under the client_lock */
1823 static void
1824 unhash_client_locked(struct nfs4_client *clp)
1825 {
1826         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1827         struct nfsd4_session *ses;
1828
1829         lockdep_assert_held(&nn->client_lock);
1830
1831         /* Mark the client as expired! */
1832         clp->cl_time = 0;
1833         /* Make it invisible */
1834         if (!list_empty(&clp->cl_idhash)) {
1835                 list_del_init(&clp->cl_idhash);
1836                 if (test_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags))
1837                         rb_erase(&clp->cl_namenode, &nn->conf_name_tree);
1838                 else
1839                         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
1840         }
1841         list_del_init(&clp->cl_lru);
1842         spin_lock(&clp->cl_lock);
1843         list_for_each_entry(ses, &clp->cl_sessions, se_perclnt)
1844                 list_del_init(&ses->se_hash);
1845         spin_unlock(&clp->cl_lock);
1846 }
1847
1848 static void
1849 unhash_client(struct nfs4_client *clp)
1850 {
1851         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
1852
1853         spin_lock(&nn->client_lock);
1854         unhash_client_locked(clp);
1855         spin_unlock(&nn->client_lock);
1856 }
1857
1858 static __be32 mark_client_expired_locked(struct nfs4_client *clp)
1859 {
1860         if (atomic_read(&clp->cl_refcount))
1861                 return nfserr_jukebox;
1862         unhash_client_locked(clp);
1863         return nfs_ok;
1864 }
1865
1866 static void
1867 __destroy_client(struct nfs4_client *clp)
1868 {
1869         struct nfs4_openowner *oo;
1870         struct nfs4_delegation *dp;
1871         struct list_head reaplist;
1872
1873         INIT_LIST_HEAD(&reaplist);
1874         spin_lock(&state_lock);
1875         while (!list_empty(&clp->cl_delegations)) {
1876                 dp = list_entry(clp->cl_delegations.next, struct nfs4_delegation, dl_perclnt);
1877                 WARN_ON(!unhash_delegation_locked(dp));
1878                 list_add(&dp->dl_recall_lru, &reaplist);
1879         }
1880         spin_unlock(&state_lock);
1881         while (!list_empty(&reaplist)) {
1882                 dp = list_entry(reaplist.next, struct nfs4_delegation, dl_recall_lru);
1883                 list_del_init(&dp->dl_recall_lru);
1884                 put_clnt_odstate(dp->dl_clnt_odstate);
1885                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
1886                 nfs4_put_stid(&dp->dl_stid);
1887         }
1888         while (!list_empty(&clp->cl_revoked)) {
1889                 dp = list_entry(clp->cl_revoked.next, struct nfs4_delegation, dl_recall_lru);
1890                 list_del_init(&dp->dl_recall_lru);
1891                 nfs4_put_stid(&dp->dl_stid);
1892         }
1893         while (!list_empty(&clp->cl_openowners)) {
1894                 oo = list_entry(clp->cl_openowners.next, struct nfs4_openowner, oo_perclient);
1895                 nfs4_get_stateowner(&oo->oo_owner);
1896                 release_openowner(oo);
1897         }
1898         nfsd4_return_all_client_layouts(clp);
1899         nfsd4_shutdown_callback(clp);
1900         if (clp->cl_cb_conn.cb_xprt)
1901                 svc_xprt_put(clp->cl_cb_conn.cb_xprt);
1902         free_client(clp);
1903 }
1904
1905 static void
1906 destroy_client(struct nfs4_client *clp)
1907 {
1908         unhash_client(clp);
1909         __destroy_client(clp);
1910 }
1911
1912 static void expire_client(struct nfs4_client *clp)
1913 {
1914         unhash_client(clp);
1915         nfsd4_client_record_remove(clp);
1916         __destroy_client(clp);
1917 }
1918
1919 static void copy_verf(struct nfs4_client *target, nfs4_verifier *source)
1920 {
1921         memcpy(target->cl_verifier.data, source->data,
1922                         sizeof(target->cl_verifier.data));
1923 }
1924
1925 static void copy_clid(struct nfs4_client *target, struct nfs4_client *source)
1926 {
1927         target->cl_clientid.cl_boot = source->cl_clientid.cl_boot;
1928         target->cl_clientid.cl_id = source->cl_clientid.cl_id;
1929 }
1930
1931 static int copy_cred(struct svc_cred *target, struct svc_cred *source)
1932 {
1933         target->cr_principal = kstrdup(source->cr_principal, GFP_KERNEL);
1934         target->cr_raw_principal = kstrdup(source->cr_raw_principal,
1935                                                                 GFP_KERNEL);
1936         if ((source->cr_principal && ! target->cr_principal) ||
1937             (source->cr_raw_principal && ! target->cr_raw_principal))
1938                 return -ENOMEM;
1939
1940         target->cr_flavor = source->cr_flavor;
1941         target->cr_uid = source->cr_uid;
1942         target->cr_gid = source->cr_gid;
1943         target->cr_group_info = source->cr_group_info;
1944         get_group_info(target->cr_group_info);
1945         target->cr_gss_mech = source->cr_gss_mech;
1946         if (source->cr_gss_mech)
1947                 gss_mech_get(source->cr_gss_mech);
1948         return 0;
1949 }
1950
1951 static int
1952 compare_blob(const struct xdr_netobj *o1, const struct xdr_netobj *o2)
1953 {
1954         if (o1->len < o2->len)
1955                 return -1;
1956         if (o1->len > o2->len)
1957                 return 1;
1958         return memcmp(o1->data, o2->data, o1->len);
1959 }
1960
1961 static int same_name(const char *n1, const char *n2)
1962 {
1963         return 0 == memcmp(n1, n2, HEXDIR_LEN);
1964 }
1965
1966 static int
1967 same_verf(nfs4_verifier *v1, nfs4_verifier *v2)
1968 {
1969         return 0 == memcmp(v1->data, v2->data, sizeof(v1->data));
1970 }
1971
1972 static int
1973 same_clid(clientid_t *cl1, clientid_t *cl2)
1974 {
1975         return (cl1->cl_boot == cl2->cl_boot) && (cl1->cl_id == cl2->cl_id);
1976 }
1977
1978 static bool groups_equal(struct group_info *g1, struct group_info *g2)
1979 {
1980         int i;
1981
1982         if (g1->ngroups != g2->ngroups)
1983                 return false;
1984         for (i=0; i<g1->ngroups; i++)
1985                 if (!gid_eq(g1->gid[i], g2->gid[i]))
1986                         return false;
1987         return true;
1988 }
1989
1990 /*
1991  * RFC 3530 language requires clid_inuse be returned when the
1992  * "principal" associated with a requests differs from that previously
1993  * used.  We use uid, gid's, and gss principal string as our best
1994  * approximation.  We also don't want to allow non-gss use of a client
1995  * established using gss: in theory cr_principal should catch that
1996  * change, but in practice cr_principal can be null even in the gss case
1997  * since gssd doesn't always pass down a principal string.
1998  */
1999 static bool is_gss_cred(struct svc_cred *cr)
2000 {
2001         /* Is cr_flavor one of the gss "pseudoflavors"?: */
2002         return (cr->cr_flavor > RPC_AUTH_MAXFLAVOR);
2003 }
2004
2005
2006 static bool
2007 same_creds(struct svc_cred *cr1, struct svc_cred *cr2)
2008 {
2009         if ((is_gss_cred(cr1) != is_gss_cred(cr2))
2010                 || (!uid_eq(cr1->cr_uid, cr2->cr_uid))
2011                 || (!gid_eq(cr1->cr_gid, cr2->cr_gid))
2012                 || !groups_equal(cr1->cr_group_info, cr2->cr_group_info))
2013                 return false;
2014         if (cr1->cr_principal == cr2->cr_principal)
2015                 return true;
2016         if (!cr1->cr_principal || !cr2->cr_principal)
2017                 return false;
2018         return 0 == strcmp(cr1->cr_principal, cr2->cr_principal);
2019 }
2020
2021 static bool svc_rqst_integrity_protected(struct svc_rqst *rqstp)
2022 {
2023         struct svc_cred *cr = &rqstp->rq_cred;
2024         u32 service;
2025
2026         if (!cr->cr_gss_mech)
2027                 return false;
2028         service = gss_pseudoflavor_to_service(cr->cr_gss_mech, cr->cr_flavor);
2029         return service == RPC_GSS_SVC_INTEGRITY ||
2030                service == RPC_GSS_SVC_PRIVACY;
2031 }
2032
2033 bool nfsd4_mach_creds_match(struct nfs4_client *cl, struct svc_rqst *rqstp)
2034 {
2035         struct svc_cred *cr = &rqstp->rq_cred;
2036
2037         if (!cl->cl_mach_cred)
2038                 return true;
2039         if (cl->cl_cred.cr_gss_mech != cr->cr_gss_mech)
2040                 return false;
2041         if (!svc_rqst_integrity_protected(rqstp))
2042                 return false;
2043         if (cl->cl_cred.cr_raw_principal)
2044                 return 0 == strcmp(cl->cl_cred.cr_raw_principal,
2045                                                 cr->cr_raw_principal);
2046         if (!cr->cr_principal)
2047                 return false;
2048         return 0 == strcmp(cl->cl_cred.cr_principal, cr->cr_principal);
2049 }
2050
2051 static void gen_confirm(struct nfs4_client *clp, struct nfsd_net *nn)
2052 {
2053         __be32 verf[2];
2054
2055         /*
2056          * This is opaque to client, so no need to byte-swap. Use
2057          * __force to keep sparse happy
2058          */
2059         verf[0] = (__force __be32)get_seconds();
2060         verf[1] = (__force __be32)nn->clverifier_counter++;
2061         memcpy(clp->cl_confirm.data, verf, sizeof(clp->cl_confirm.data));
2062 }
2063
2064 static void gen_clid(struct nfs4_client *clp, struct nfsd_net *nn)
2065 {
2066         clp->cl_clientid.cl_boot = nn->boot_time;
2067         clp->cl_clientid.cl_id = nn->clientid_counter++;
2068         gen_confirm(clp, nn);
2069 }
2070
2071 static struct nfs4_stid *
2072 find_stateid_locked(struct nfs4_client *cl, stateid_t *t)
2073 {
2074         struct nfs4_stid *ret;
2075
2076         ret = idr_find(&cl->cl_stateids, t->si_opaque.so_id);
2077         if (!ret || !ret->sc_type)
2078                 return NULL;
2079         return ret;
2080 }
2081
2082 static struct nfs4_stid *
2083 find_stateid_by_type(struct nfs4_client *cl, stateid_t *t, char typemask)
2084 {
2085         struct nfs4_stid *s;
2086
2087         spin_lock(&cl->cl_lock);
2088         s = find_stateid_locked(cl, t);
2089         if (s != NULL) {
2090                 if (typemask & s->sc_type)
2091                         refcount_inc(&s->sc_count);
2092                 else
2093                         s = NULL;
2094         }
2095         spin_unlock(&cl->cl_lock);
2096         return s;
2097 }
2098
2099 static struct nfs4_client *create_client(struct xdr_netobj name,
2100                 struct svc_rqst *rqstp, nfs4_verifier *verf)
2101 {
2102         struct nfs4_client *clp;
2103         struct sockaddr *sa = svc_addr(rqstp);
2104         int ret;
2105         struct net *net = SVC_NET(rqstp);
2106
2107         clp = alloc_client(name);
2108         if (clp == NULL)
2109                 return NULL;
2110
2111         ret = copy_cred(&clp->cl_cred, &rqstp->rq_cred);
2112         if (ret) {
2113                 free_client(clp);
2114                 return NULL;
2115         }
2116         nfsd4_init_cb(&clp->cl_cb_null, clp, NULL, NFSPROC4_CLNT_CB_NULL);
2117         clp->cl_time = get_seconds();
2118         clear_bit(0, &clp->cl_cb_slot_busy);
2119         copy_verf(clp, verf);
2120         rpc_copy_addr((struct sockaddr *) &clp->cl_addr, sa);
2121         clp->cl_cb_session = NULL;
2122         clp->net = net;
2123         return clp;
2124 }
2125
2126 static void
2127 add_clp_to_name_tree(struct nfs4_client *new_clp, struct rb_root *root)
2128 {
2129         struct rb_node **new = &(root->rb_node), *parent = NULL;
2130         struct nfs4_client *clp;
2131
2132         while (*new) {
2133                 clp = rb_entry(*new, struct nfs4_client, cl_namenode);
2134                 parent = *new;
2135
2136                 if (compare_blob(&clp->cl_name, &new_clp->cl_name) > 0)
2137                         new = &((*new)->rb_left);
2138                 else
2139                         new = &((*new)->rb_right);
2140         }
2141
2142         rb_link_node(&new_clp->cl_namenode, parent, new);
2143         rb_insert_color(&new_clp->cl_namenode, root);
2144 }
2145
2146 static struct nfs4_client *
2147 find_clp_in_name_tree(struct xdr_netobj *name, struct rb_root *root)
2148 {
2149         int cmp;
2150         struct rb_node *node = root->rb_node;
2151         struct nfs4_client *clp;
2152
2153         while (node) {
2154                 clp = rb_entry(node, struct nfs4_client, cl_namenode);
2155                 cmp = compare_blob(&clp->cl_name, name);
2156                 if (cmp > 0)
2157                         node = node->rb_left;
2158                 else if (cmp < 0)
2159                         node = node->rb_right;
2160                 else
2161                         return clp;
2162         }
2163         return NULL;
2164 }
2165
2166 static void
2167 add_to_unconfirmed(struct nfs4_client *clp)
2168 {
2169         unsigned int idhashval;
2170         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2171
2172         lockdep_assert_held(&nn->client_lock);
2173
2174         clear_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2175         add_clp_to_name_tree(clp, &nn->unconf_name_tree);
2176         idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2177         list_add(&clp->cl_idhash, &nn->unconf_id_hashtbl[idhashval]);
2178         renew_client_locked(clp);
2179 }
2180
2181 static void
2182 move_to_confirmed(struct nfs4_client *clp)
2183 {
2184         unsigned int idhashval = clientid_hashval(clp->cl_clientid.cl_id);
2185         struct nfsd_net *nn = net_generic(clp->net, nfsd_net_id);
2186
2187         lockdep_assert_held(&nn->client_lock);
2188
2189         dprintk("NFSD: move_to_confirm nfs4_client %p\n", clp);
2190         list_move(&clp->cl_idhash, &nn->conf_id_hashtbl[idhashval]);
2191         rb_erase(&clp->cl_namenode, &nn->unconf_name_tree);
2192         add_clp_to_name_tree(clp, &nn->conf_name_tree);
2193         set_bit(NFSD4_CLIENT_CONFIRMED, &clp->cl_flags);
2194         renew_client_locked(clp);
2195 }
2196
2197 static struct nfs4_client *
2198 find_client_in_id_table(struct list_head *tbl, clientid_t *clid, bool sessions)
2199 {
2200         struct nfs4_client *clp;
2201         unsigned int idhashval = clientid_hashval(clid->cl_id);
2202
2203         list_for_each_entry(clp, &tbl[idhashval], cl_idhash) {
2204                 if (same_clid(&clp->cl_clientid, clid)) {
2205                         if ((bool)clp->cl_minorversion != sessions)
2206                                 return NULL;
2207                         renew_client_locked(clp);
2208                         return clp;
2209                 }
2210         }
2211         return NULL;
2212 }
2213
2214 static struct nfs4_client *
2215 find_confirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2216 {
2217         struct list_head *tbl = nn->conf_id_hashtbl;
2218
2219         lockdep_assert_held(&nn->client_lock);
2220         return find_client_in_id_table(tbl, clid, sessions);
2221 }
2222
2223 static struct nfs4_client *
2224 find_unconfirmed_client(clientid_t *clid, bool sessions, struct nfsd_net *nn)
2225 {
2226         struct list_head *tbl = nn->unconf_id_hashtbl;
2227
2228         lockdep_assert_held(&nn->client_lock);
2229         return find_client_in_id_table(tbl, clid, sessions);
2230 }
2231
2232 static bool clp_used_exchangeid(struct nfs4_client *clp)
2233 {
2234         return clp->cl_exchange_flags != 0;
2235 }
2236
2237 static struct nfs4_client *
2238 find_confirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2239 {
2240         lockdep_assert_held(&nn->client_lock);
2241         return find_clp_in_name_tree(name, &nn->conf_name_tree);
2242 }
2243
2244 static struct nfs4_client *
2245 find_unconfirmed_client_by_name(struct xdr_netobj *name, struct nfsd_net *nn)
2246 {
2247         lockdep_assert_held(&nn->client_lock);
2248         return find_clp_in_name_tree(name, &nn->unconf_name_tree);
2249 }
2250
2251 static void
2252 gen_callback(struct nfs4_client *clp, struct nfsd4_setclientid *se, struct svc_rqst *rqstp)
2253 {
2254         struct nfs4_cb_conn *conn = &clp->cl_cb_conn;
2255         struct sockaddr *sa = svc_addr(rqstp);
2256         u32 scopeid = rpc_get_scope_id(sa);
2257         unsigned short expected_family;
2258
2259         /* Currently, we only support tcp and tcp6 for the callback channel */
2260         if (se->se_callback_netid_len == 3 &&
2261             !memcmp(se->se_callback_netid_val, "tcp", 3))
2262                 expected_family = AF_INET;
2263         else if (se->se_callback_netid_len == 4 &&
2264                  !memcmp(se->se_callback_netid_val, "tcp6", 4))
2265                 expected_family = AF_INET6;
2266         else
2267                 goto out_err;
2268
2269         conn->cb_addrlen = rpc_uaddr2sockaddr(clp->net, se->se_callback_addr_val,
2270                                             se->se_callback_addr_len,
2271                                             (struct sockaddr *)&conn->cb_addr,
2272                                             sizeof(conn->cb_addr));
2273
2274         if (!conn->cb_addrlen || conn->cb_addr.ss_family != expected_family)
2275                 goto out_err;
2276
2277         if (conn->cb_addr.ss_family == AF_INET6)
2278                 ((struct sockaddr_in6 *)&conn->cb_addr)->sin6_scope_id = scopeid;
2279
2280         conn->cb_prog = se->se_callback_prog;
2281         conn->cb_ident = se->se_callback_ident;
2282         memcpy(&conn->cb_saddr, &rqstp->rq_daddr, rqstp->rq_daddrlen);
2283         return;
2284 out_err:
2285         conn->cb_addr.ss_family = AF_UNSPEC;
2286         conn->cb_addrlen = 0;
2287         dprintk("NFSD: this client (clientid %08x/%08x) "
2288                 "will not receive delegations\n",
2289                 clp->cl_clientid.cl_boot, clp->cl_clientid.cl_id);
2290
2291         return;
2292 }
2293
2294 /*
2295  * Cache a reply. nfsd4_check_resp_size() has bounded the cache size.
2296  */
2297 static void
2298 nfsd4_store_cache_entry(struct nfsd4_compoundres *resp)
2299 {
2300         struct xdr_buf *buf = resp->xdr.buf;
2301         struct nfsd4_slot *slot = resp->cstate.slot;
2302         unsigned int base;
2303
2304         dprintk("--> %s slot %p\n", __func__, slot);
2305
2306         slot->sl_flags |= NFSD4_SLOT_INITIALIZED;
2307         slot->sl_opcnt = resp->opcnt;
2308         slot->sl_status = resp->cstate.status;
2309         free_svc_cred(&slot->sl_cred);
2310         copy_cred(&slot->sl_cred, &resp->rqstp->rq_cred);
2311
2312         if (!nfsd4_cache_this(resp)) {
2313                 slot->sl_flags &= ~NFSD4_SLOT_CACHED;
2314                 return;
2315         }
2316         slot->sl_flags |= NFSD4_SLOT_CACHED;
2317
2318         base = resp->cstate.data_offset;
2319         slot->sl_datalen = buf->len - base;
2320         if (read_bytes_from_xdr_buf(buf, base, slot->sl_data, slot->sl_datalen))
2321                 WARN(1, "%s: sessions DRC could not cache compound\n",
2322                      __func__);
2323         return;
2324 }
2325
2326 /*
2327  * Encode the replay sequence operation from the slot values.
2328  * If cachethis is FALSE encode the uncached rep error on the next
2329  * operation which sets resp->p and increments resp->opcnt for
2330  * nfs4svc_encode_compoundres.
2331  *
2332  */
2333 static __be32
2334 nfsd4_enc_sequence_replay(struct nfsd4_compoundargs *args,
2335                           struct nfsd4_compoundres *resp)
2336 {
2337         struct nfsd4_op *op;
2338         struct nfsd4_slot *slot = resp->cstate.slot;
2339
2340         /* Encode the replayed sequence operation */
2341         op = &args->ops[resp->opcnt - 1];
2342         nfsd4_encode_operation(resp, op);
2343
2344         if (slot->sl_flags & NFSD4_SLOT_CACHED)
2345                 return op->status;
2346         if (args->opcnt == 1) {
2347                 /*
2348                  * The original operation wasn't a solo sequence--we
2349                  * always cache those--so this retry must not match the
2350                  * original:
2351                  */
2352                 op->status = nfserr_seq_false_retry;
2353         } else {
2354                 op = &args->ops[resp->opcnt++];
2355                 op->status = nfserr_retry_uncached_rep;
2356                 nfsd4_encode_operation(resp, op);
2357         }
2358         return op->status;
2359 }
2360
2361 /*
2362  * The sequence operation is not cached because we can use the slot and
2363  * session values.
2364  */
2365 static __be32
2366 nfsd4_replay_cache_entry(struct nfsd4_compoundres *resp,
2367                          struct nfsd4_sequence *seq)
2368 {
2369         struct nfsd4_slot *slot = resp->cstate.slot;
2370         struct xdr_stream *xdr = &resp->xdr;
2371         __be32 *p;
2372         __be32 status;
2373
2374         dprintk("--> %s slot %p\n", __func__, slot);
2375
2376         status = nfsd4_enc_sequence_replay(resp->rqstp->rq_argp, resp);
2377         if (status)
2378                 return status;
2379
2380         p = xdr_reserve_space(xdr, slot->sl_datalen);
2381         if (!p) {
2382                 WARN_ON_ONCE(1);
2383                 return nfserr_serverfault;
2384         }
2385         xdr_encode_opaque_fixed(p, slot->sl_data, slot->sl_datalen);
2386         xdr_commit_encode(xdr);
2387
2388         resp->opcnt = slot->sl_opcnt;
2389         return slot->sl_status;
2390 }
2391
2392 /*
2393  * Set the exchange_id flags returned by the server.
2394  */
2395 static void
2396 nfsd4_set_ex_flags(struct nfs4_client *new, struct nfsd4_exchange_id *clid)
2397 {
2398 #ifdef CONFIG_NFSD_PNFS
2399         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_PNFS_MDS;
2400 #else
2401         new->cl_exchange_flags |= EXCHGID4_FLAG_USE_NON_PNFS;
2402 #endif
2403
2404         /* Referrals are supported, Migration is not. */
2405         new->cl_exchange_flags |= EXCHGID4_FLAG_SUPP_MOVED_REFER;
2406
2407         /* set the wire flags to return to client. */
2408         clid->flags = new->cl_exchange_flags;
2409 }
2410
2411 static bool client_has_openowners(struct nfs4_client *clp)
2412 {
2413         struct nfs4_openowner *oo;
2414
2415         list_for_each_entry(oo, &clp->cl_openowners, oo_perclient) {
2416                 if (!list_empty(&oo->oo_owner.so_stateids))
2417                         return true;
2418         }
2419         return false;
2420 }
2421
2422 static bool client_has_state(struct nfs4_client *clp)
2423 {
2424         return client_has_openowners(clp)
2425 #ifdef CONFIG_NFSD_PNFS
2426                 || !list_empty(&clp->cl_lo_states)
2427 #endif
2428                 || !list_empty(&clp->cl_delegations)
2429                 || !list_empty(&clp->cl_sessions);
2430 }
2431
2432 __be32
2433 nfsd4_exchange_id(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
2434                 union nfsd4_op_u *u)
2435 {
2436         struct nfsd4_exchange_id *exid = &u->exchange_id;
2437         struct nfs4_client *conf, *new;
2438         struct nfs4_client *unconf = NULL;
2439         __be32 status;
2440         char                    addr_str[INET6_ADDRSTRLEN];
2441         nfs4_verifier           verf = exid->verifier;
2442         struct sockaddr         *sa = svc_addr(rqstp);
2443         bool    update = exid->flags & EXCHGID4_FLAG_UPD_CONFIRMED_REC_A;
2444         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2445
2446         rpc_ntop(sa, addr_str, sizeof(addr_str));
2447         dprintk("%s rqstp=%p exid=%p clname.len=%u clname.data=%p "
2448                 "ip_addr=%s flags %x, spa_how %d\n",
2449                 __func__, rqstp, exid, exid->clname.len, exid->clname.data,
2450                 addr_str, exid->flags, exid->spa_how);
2451
2452         if (exid->flags & ~EXCHGID4_FLAG_MASK_A)
2453                 return nfserr_inval;
2454
2455         new = create_client(exid->clname, rqstp, &verf);
2456         if (new == NULL)
2457                 return nfserr_jukebox;
2458
2459         switch (exid->spa_how) {
2460         case SP4_MACH_CRED:
2461                 exid->spo_must_enforce[0] = 0;
2462                 exid->spo_must_enforce[1] = (
2463                         1 << (OP_BIND_CONN_TO_SESSION - 32) |
2464                         1 << (OP_EXCHANGE_ID - 32) |
2465                         1 << (OP_CREATE_SESSION - 32) |
2466                         1 << (OP_DESTROY_SESSION - 32) |
2467                         1 << (OP_DESTROY_CLIENTID - 32));
2468
2469                 exid->spo_must_allow[0] &= (1 << (OP_CLOSE) |
2470                                         1 << (OP_OPEN_DOWNGRADE) |
2471                                         1 << (OP_LOCKU) |
2472                                         1 << (OP_DELEGRETURN));
2473
2474                 exid->spo_must_allow[1] &= (
2475                                         1 << (OP_TEST_STATEID - 32) |
2476                                         1 << (OP_FREE_STATEID - 32));
2477                 if (!svc_rqst_integrity_protected(rqstp)) {
2478                         status = nfserr_inval;
2479                         goto out_nolock;
2480                 }
2481                 /*
2482                  * Sometimes userspace doesn't give us a principal.
2483                  * Which is a bug, really.  Anyway, we can't enforce
2484                  * MACH_CRED in that case, better to give up now:
2485                  */
2486                 if (!new->cl_cred.cr_principal &&
2487                                         !new->cl_cred.cr_raw_principal) {
2488                         status = nfserr_serverfault;
2489                         goto out_nolock;
2490                 }
2491                 new->cl_mach_cred = true;
2492         case SP4_NONE:
2493                 break;
2494         default:                                /* checked by xdr code */
2495                 WARN_ON_ONCE(1);
2496         case SP4_SSV:
2497                 status = nfserr_encr_alg_unsupp;
2498                 goto out_nolock;
2499         }
2500
2501         /* Cases below refer to rfc 5661 section 18.35.4: */
2502         spin_lock(&nn->client_lock);
2503         conf = find_confirmed_client_by_name(&exid->clname, nn);
2504         if (conf) {
2505                 bool creds_match = same_creds(&conf->cl_cred, &rqstp->rq_cred);
2506                 bool verfs_match = same_verf(&verf, &conf->cl_verifier);
2507
2508                 if (update) {
2509                         if (!clp_used_exchangeid(conf)) { /* buggy client */
2510                                 status = nfserr_inval;
2511                                 goto out;
2512                         }
2513                         if (!nfsd4_mach_creds_match(conf, rqstp)) {
2514                                 status = nfserr_wrong_cred;
2515                                 goto out;
2516                         }
2517                         if (!creds_match) { /* case 9 */
2518                                 status = nfserr_perm;
2519                                 goto out;
2520                         }
2521                         if (!verfs_match) { /* case 8 */
2522                                 status = nfserr_not_same;
2523                                 goto out;
2524                         }
2525                         /* case 6 */
2526                         exid->flags |= EXCHGID4_FLAG_CONFIRMED_R;
2527                         goto out_copy;
2528                 }
2529                 if (!creds_match) { /* case 3 */
2530                         if (client_has_state(conf)) {
2531                                 status = nfserr_clid_inuse;
2532                                 goto out;
2533                         }
2534                         goto out_new;
2535                 }
2536                 if (verfs_match) { /* case 2 */
2537                         conf->cl_exchange_flags |= EXCHGID4_FLAG_CONFIRMED_R;
2538                         goto out_copy;
2539                 }
2540                 /* case 5, client reboot */
2541                 conf = NULL;
2542                 goto out_new;
2543         }
2544
2545         if (update) { /* case 7 */
2546                 status = nfserr_noent;
2547                 goto out;
2548         }
2549
2550         unconf  = find_unconfirmed_client_by_name(&exid->clname, nn);
2551         if (unconf) /* case 4, possible retry or client restart */
2552                 unhash_client_locked(unconf);
2553
2554         /* case 1 (normal case) */
2555 out_new:
2556         if (conf) {
2557                 status = mark_client_expired_locked(conf);
2558                 if (status)
2559                         goto out;
2560         }
2561         new->cl_minorversion = cstate->minorversion;
2562         new->cl_spo_must_allow.u.words[0] = exid->spo_must_allow[0];
2563         new->cl_spo_must_allow.u.words[1] = exid->spo_must_allow[1];
2564
2565         gen_clid(new, nn);
2566         add_to_unconfirmed(new);
2567         swap(new, conf);
2568 out_copy:
2569         exid->clientid.cl_boot = conf->cl_clientid.cl_boot;
2570         exid->clientid.cl_id = conf->cl_clientid.cl_id;
2571
2572         exid->seqid = conf->cl_cs_slot.sl_seqid + 1;
2573         nfsd4_set_ex_flags(conf, exid);
2574
2575         dprintk("nfsd4_exchange_id seqid %d flags %x\n",
2576                 conf->cl_cs_slot.sl_seqid, conf->cl_exchange_flags);
2577         status = nfs_ok;
2578
2579 out:
2580         spin_unlock(&nn->client_lock);
2581 out_nolock:
2582         if (new)
2583                 expire_client(new);
2584         if (unconf)
2585                 expire_client(unconf);
2586         return status;
2587 }
2588
2589 static __be32
2590 check_slot_seqid(u32 seqid, u32 slot_seqid, int slot_inuse)
2591 {
2592         dprintk("%s enter. seqid %d slot_seqid %d\n", __func__, seqid,
2593                 slot_seqid);
2594
2595         /* The slot is in use, and no response has been sent. */
2596         if (slot_inuse) {
2597                 if (seqid == slot_seqid)
2598                         return nfserr_jukebox;
2599                 else
2600                         return nfserr_seq_misordered;
2601         }
2602         /* Note unsigned 32-bit arithmetic handles wraparound: */
2603         if (likely(seqid == slot_seqid + 1))
2604                 return nfs_ok;
2605         if (seqid == slot_seqid)
2606                 return nfserr_replay_cache;
2607         return nfserr_seq_misordered;
2608 }
2609
2610 /*
2611  * Cache the create session result into the create session single DRC
2612  * slot cache by saving the xdr structure. sl_seqid has been set.
2613  * Do this for solo or embedded create session operations.
2614  */
2615 static void
2616 nfsd4_cache_create_session(struct nfsd4_create_session *cr_ses,
2617                            struct nfsd4_clid_slot *slot, __be32 nfserr)
2618 {
2619         slot->sl_status = nfserr;
2620         memcpy(&slot->sl_cr_ses, cr_ses, sizeof(*cr_ses));
2621 }
2622
2623 static __be32
2624 nfsd4_replay_create_session(struct nfsd4_create_session *cr_ses,
2625                             struct nfsd4_clid_slot *slot)
2626 {
2627         memcpy(cr_ses, &slot->sl_cr_ses, sizeof(*cr_ses));
2628         return slot->sl_status;
2629 }
2630
2631 #define NFSD_MIN_REQ_HDR_SEQ_SZ ((\
2632                         2 * 2 + /* credential,verifier: AUTH_NULL, length 0 */ \
2633                         1 +     /* MIN tag is length with zero, only length */ \
2634                         3 +     /* version, opcount, opcode */ \
2635                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2636                                 /* seqid, slotID, slotID, cache */ \
2637                         4 ) * sizeof(__be32))
2638
2639 #define NFSD_MIN_RESP_HDR_SEQ_SZ ((\
2640                         2 +     /* verifier: AUTH_NULL, length 0 */\
2641                         1 +     /* status */ \
2642                         1 +     /* MIN tag is length with zero, only length */ \
2643                         3 +     /* opcount, opcode, opstatus*/ \
2644                         XDR_QUADLEN(NFS4_MAX_SESSIONID_LEN) + \
2645                                 /* seqid, slotID, slotID, slotID, status */ \
2646                         5 ) * sizeof(__be32))
2647
2648 static __be32 check_forechannel_attrs(struct nfsd4_channel_attrs *ca, struct nfsd_net *nn)
2649 {
2650         u32 maxrpc = nn->nfsd_serv->sv_max_mesg;
2651
2652         if (ca->maxreq_sz < NFSD_MIN_REQ_HDR_SEQ_SZ)
2653                 return nfserr_toosmall;
2654         if (ca->maxresp_sz < NFSD_MIN_RESP_HDR_SEQ_SZ)
2655                 return nfserr_toosmall;
2656         ca->headerpadsz = 0;
2657         ca->maxreq_sz = min_t(u32, ca->maxreq_sz, maxrpc);
2658         ca->maxresp_sz = min_t(u32, ca->maxresp_sz, maxrpc);
2659         ca->maxops = min_t(u32, ca->maxops, NFSD_MAX_OPS_PER_COMPOUND);
2660         ca->maxresp_cached = min_t(u32, ca->maxresp_cached,
2661                         NFSD_SLOT_CACHE_SIZE + NFSD_MIN_HDR_SEQ_SZ);
2662         ca->maxreqs = min_t(u32, ca->maxreqs, NFSD_MAX_SLOTS_PER_SESSION);
2663         /*
2664          * Note decreasing slot size below client's request may make it
2665          * difficult for client to function correctly, whereas
2666          * decreasing the number of slots will (just?) affect
2667          * performance.  When short on memory we therefore prefer to
2668          * decrease number of slots instead of their size.  Clients that
2669          * request larger slots than they need will get poor results:
2670          */
2671         ca->maxreqs = nfsd4_get_drc_mem(ca);
2672         if (!ca->maxreqs)
2673                 return nfserr_jukebox;
2674
2675         return nfs_ok;
2676 }
2677
2678 /*
2679  * Server's NFSv4.1 backchannel support is AUTH_SYS-only for now.
2680  * These are based on similar macros in linux/sunrpc/msg_prot.h .
2681  */
2682 #define RPC_MAX_HEADER_WITH_AUTH_SYS \
2683         (RPC_CALLHDRSIZE + 2 * (2 + UNX_CALLSLACK))
2684
2685 #define RPC_MAX_REPHEADER_WITH_AUTH_SYS \
2686         (RPC_REPHDRSIZE + (2 + NUL_REPLYSLACK))
2687
2688 #define NFSD_CB_MAX_REQ_SZ      ((NFS4_enc_cb_recall_sz + \
2689                                  RPC_MAX_HEADER_WITH_AUTH_SYS) * sizeof(__be32))
2690 #define NFSD_CB_MAX_RESP_SZ     ((NFS4_dec_cb_recall_sz + \
2691                                  RPC_MAX_REPHEADER_WITH_AUTH_SYS) * \
2692                                  sizeof(__be32))
2693
2694 static __be32 check_backchannel_attrs(struct nfsd4_channel_attrs *ca)
2695 {
2696         ca->headerpadsz = 0;
2697
2698         if (ca->maxreq_sz < NFSD_CB_MAX_REQ_SZ)
2699                 return nfserr_toosmall;
2700         if (ca->maxresp_sz < NFSD_CB_MAX_RESP_SZ)
2701                 return nfserr_toosmall;
2702         ca->maxresp_cached = 0;
2703         if (ca->maxops < 2)
2704                 return nfserr_toosmall;
2705
2706         return nfs_ok;
2707 }
2708
2709 static __be32 nfsd4_check_cb_sec(struct nfsd4_cb_sec *cbs)
2710 {
2711         switch (cbs->flavor) {
2712         case RPC_AUTH_NULL:
2713         case RPC_AUTH_UNIX:
2714                 return nfs_ok;
2715         default:
2716                 /*
2717                  * GSS case: the spec doesn't allow us to return this
2718                  * error.  But it also doesn't allow us not to support
2719                  * GSS.
2720                  * I'd rather this fail hard than return some error the
2721                  * client might think it can already handle:
2722                  */
2723                 return nfserr_encr_alg_unsupp;
2724         }
2725 }
2726
2727 __be32
2728 nfsd4_create_session(struct svc_rqst *rqstp,
2729                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
2730 {
2731         struct nfsd4_create_session *cr_ses = &u->create_session;
2732         struct sockaddr *sa = svc_addr(rqstp);
2733         struct nfs4_client *conf, *unconf;
2734         struct nfs4_client *old = NULL;
2735         struct nfsd4_session *new;
2736         struct nfsd4_conn *conn;
2737         struct nfsd4_clid_slot *cs_slot = NULL;
2738         __be32 status = 0;
2739         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2740
2741         if (cr_ses->flags & ~SESSION4_FLAG_MASK_A)
2742                 return nfserr_inval;
2743         status = nfsd4_check_cb_sec(&cr_ses->cb_sec);
2744         if (status)
2745                 return status;
2746         status = check_forechannel_attrs(&cr_ses->fore_channel, nn);
2747         if (status)
2748                 return status;
2749         status = check_backchannel_attrs(&cr_ses->back_channel);
2750         if (status)
2751                 goto out_release_drc_mem;
2752         status = nfserr_jukebox;
2753         new = alloc_session(&cr_ses->fore_channel, &cr_ses->back_channel);
2754         if (!new)
2755                 goto out_release_drc_mem;
2756         conn = alloc_conn_from_crses(rqstp, cr_ses);
2757         if (!conn)
2758                 goto out_free_session;
2759
2760         spin_lock(&nn->client_lock);
2761         unconf = find_unconfirmed_client(&cr_ses->clientid, true, nn);
2762         conf = find_confirmed_client(&cr_ses->clientid, true, nn);
2763         WARN_ON_ONCE(conf && unconf);
2764
2765         if (conf) {
2766                 status = nfserr_wrong_cred;
2767                 if (!nfsd4_mach_creds_match(conf, rqstp))
2768                         goto out_free_conn;
2769                 cs_slot = &conf->cl_cs_slot;
2770                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2771                 if (status) {
2772                         if (status == nfserr_replay_cache)
2773                                 status = nfsd4_replay_create_session(cr_ses, cs_slot);
2774                         goto out_free_conn;
2775                 }
2776         } else if (unconf) {
2777                 if (!same_creds(&unconf->cl_cred, &rqstp->rq_cred) ||
2778                     !rpc_cmp_addr(sa, (struct sockaddr *) &unconf->cl_addr)) {
2779                         status = nfserr_clid_inuse;
2780                         goto out_free_conn;
2781                 }
2782                 status = nfserr_wrong_cred;
2783                 if (!nfsd4_mach_creds_match(unconf, rqstp))
2784                         goto out_free_conn;
2785                 cs_slot = &unconf->cl_cs_slot;
2786                 status = check_slot_seqid(cr_ses->seqid, cs_slot->sl_seqid, 0);
2787                 if (status) {
2788                         /* an unconfirmed replay returns misordered */
2789                         status = nfserr_seq_misordered;
2790                         goto out_free_conn;
2791                 }
2792                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
2793                 if (old) {
2794                         status = mark_client_expired_locked(old);
2795                         if (status) {
2796                                 old = NULL;
2797                                 goto out_free_conn;
2798                         }
2799                 }
2800                 move_to_confirmed(unconf);
2801                 conf = unconf;
2802         } else {
2803                 status = nfserr_stale_clientid;
2804                 goto out_free_conn;
2805         }
2806         status = nfs_ok;
2807         /* Persistent sessions are not supported */
2808         cr_ses->flags &= ~SESSION4_PERSIST;
2809         /* Upshifting from TCP to RDMA is not supported */
2810         cr_ses->flags &= ~SESSION4_RDMA;
2811
2812         init_session(rqstp, new, conf, cr_ses);
2813         nfsd4_get_session_locked(new);
2814
2815         memcpy(cr_ses->sessionid.data, new->se_sessionid.data,
2816                NFS4_MAX_SESSIONID_LEN);
2817         cs_slot->sl_seqid++;
2818         cr_ses->seqid = cs_slot->sl_seqid;
2819
2820         /* cache solo and embedded create sessions under the client_lock */
2821         nfsd4_cache_create_session(cr_ses, cs_slot, status);
2822         spin_unlock(&nn->client_lock);
2823         /* init connection and backchannel */
2824         nfsd4_init_conn(rqstp, conn, new);
2825         nfsd4_put_session(new);
2826         if (old)
2827                 expire_client(old);
2828         return status;
2829 out_free_conn:
2830         spin_unlock(&nn->client_lock);
2831         free_conn(conn);
2832         if (old)
2833                 expire_client(old);
2834 out_free_session:
2835         __free_session(new);
2836 out_release_drc_mem:
2837         nfsd4_put_drc_mem(&cr_ses->fore_channel);
2838         return status;
2839 }
2840
2841 static __be32 nfsd4_map_bcts_dir(u32 *dir)
2842 {
2843         switch (*dir) {
2844         case NFS4_CDFC4_FORE:
2845         case NFS4_CDFC4_BACK:
2846                 return nfs_ok;
2847         case NFS4_CDFC4_FORE_OR_BOTH:
2848         case NFS4_CDFC4_BACK_OR_BOTH:
2849                 *dir = NFS4_CDFC4_BOTH;
2850                 return nfs_ok;
2851         };
2852         return nfserr_inval;
2853 }
2854
2855 __be32 nfsd4_backchannel_ctl(struct svc_rqst *rqstp,
2856                 struct nfsd4_compound_state *cstate,
2857                 union nfsd4_op_u *u)
2858 {
2859         struct nfsd4_backchannel_ctl *bc = &u->backchannel_ctl;
2860         struct nfsd4_session *session = cstate->session;
2861         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
2862         __be32 status;
2863
2864         status = nfsd4_check_cb_sec(&bc->bc_cb_sec);
2865         if (status)
2866                 return status;
2867         spin_lock(&nn->client_lock);
2868         session->se_cb_prog = bc->bc_cb_program;
2869         session->se_cb_sec = bc->bc_cb_sec;
2870         spin_unlock(&nn->client_lock);
2871
2872         nfsd4_probe_callback(session->se_client);
2873
2874         return nfs_ok;
2875 }
2876
2877 __be32 nfsd4_bind_conn_to_session(struct svc_rqst *rqstp,
2878                      struct nfsd4_compound_state *cstate,
2879                      union nfsd4_op_u *u)
2880 {
2881         struct nfsd4_bind_conn_to_session *bcts = &u->bind_conn_to_session;
2882         __be32 status;
2883         struct nfsd4_conn *conn;
2884         struct nfsd4_session *session;
2885         struct net *net = SVC_NET(rqstp);
2886         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2887
2888         if (!nfsd4_last_compound_op(rqstp))
2889                 return nfserr_not_only_op;
2890         spin_lock(&nn->client_lock);
2891         session = find_in_sessionid_hashtbl(&bcts->sessionid, net, &status);
2892         spin_unlock(&nn->client_lock);
2893         if (!session)
2894                 goto out_no_session;
2895         status = nfserr_wrong_cred;
2896         if (!nfsd4_mach_creds_match(session->se_client, rqstp))
2897                 goto out;
2898         status = nfsd4_map_bcts_dir(&bcts->dir);
2899         if (status)
2900                 goto out;
2901         conn = alloc_conn(rqstp, bcts->dir);
2902         status = nfserr_jukebox;
2903         if (!conn)
2904                 goto out;
2905         nfsd4_init_conn(rqstp, conn, session);
2906         status = nfs_ok;
2907 out:
2908         nfsd4_put_session(session);
2909 out_no_session:
2910         return status;
2911 }
2912
2913 static bool nfsd4_compound_in_session(struct nfsd4_session *session, struct nfs4_sessionid *sid)
2914 {
2915         if (!session)
2916                 return 0;
2917         return !memcmp(sid, &session->se_sessionid, sizeof(*sid));
2918 }
2919
2920 __be32
2921 nfsd4_destroy_session(struct svc_rqst *r, struct nfsd4_compound_state *cstate,
2922                 union nfsd4_op_u *u)
2923 {
2924         struct nfsd4_destroy_session *sessionid = &u->destroy_session;
2925         struct nfsd4_session *ses;
2926         __be32 status;
2927         int ref_held_by_me = 0;
2928         struct net *net = SVC_NET(r);
2929         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
2930
2931         status = nfserr_not_only_op;
2932         if (nfsd4_compound_in_session(cstate->session, &sessionid->sessionid)) {
2933                 if (!nfsd4_last_compound_op(r))
2934                         goto out;
2935                 ref_held_by_me++;
2936         }
2937         dump_sessionid(__func__, &sessionid->sessionid);
2938         spin_lock(&nn->client_lock);
2939         ses = find_in_sessionid_hashtbl(&sessionid->sessionid, net, &status);
2940         if (!ses)
2941                 goto out_client_lock;
2942         status = nfserr_wrong_cred;
2943         if (!nfsd4_mach_creds_match(ses->se_client, r))
2944                 goto out_put_session;
2945         status = mark_session_dead_locked(ses, 1 + ref_held_by_me);
2946         if (status)
2947                 goto out_put_session;
2948         unhash_session(ses);
2949         spin_unlock(&nn->client_lock);
2950
2951         nfsd4_probe_callback_sync(ses->se_client);
2952
2953         spin_lock(&nn->client_lock);
2954         status = nfs_ok;
2955 out_put_session:
2956         nfsd4_put_session_locked(ses);
2957 out_client_lock:
2958         spin_unlock(&nn->client_lock);
2959 out:
2960         return status;
2961 }
2962
2963 static struct nfsd4_conn *__nfsd4_find_conn(struct svc_xprt *xpt, struct nfsd4_session *s)
2964 {
2965         struct nfsd4_conn *c;
2966
2967         list_for_each_entry(c, &s->se_conns, cn_persession) {
2968                 if (c->cn_xprt == xpt) {
2969                         return c;
2970                 }
2971         }
2972         return NULL;
2973 }
2974
2975 static __be32 nfsd4_sequence_check_conn(struct nfsd4_conn *new, struct nfsd4_session *ses)
2976 {
2977         struct nfs4_client *clp = ses->se_client;
2978         struct nfsd4_conn *c;
2979         __be32 status = nfs_ok;
2980         int ret;
2981
2982         spin_lock(&clp->cl_lock);
2983         c = __nfsd4_find_conn(new->cn_xprt, ses);
2984         if (c)
2985                 goto out_free;
2986         status = nfserr_conn_not_bound_to_session;
2987         if (clp->cl_mach_cred)
2988                 goto out_free;
2989         __nfsd4_hash_conn(new, ses);
2990         spin_unlock(&clp->cl_lock);
2991         ret = nfsd4_register_conn(new);
2992         if (ret)
2993                 /* oops; xprt is already down: */
2994                 nfsd4_conn_lost(&new->cn_xpt_user);
2995         return nfs_ok;
2996 out_free:
2997         spin_unlock(&clp->cl_lock);
2998         free_conn(new);
2999         return status;
3000 }
3001
3002 static bool nfsd4_session_too_many_ops(struct svc_rqst *rqstp, struct nfsd4_session *session)
3003 {
3004         struct nfsd4_compoundargs *args = rqstp->rq_argp;
3005
3006         return args->opcnt > session->se_fchannel.maxops;
3007 }
3008
3009 static bool nfsd4_request_too_big(struct svc_rqst *rqstp,
3010                                   struct nfsd4_session *session)
3011 {
3012         struct xdr_buf *xb = &rqstp->rq_arg;
3013
3014         return xb->len > session->se_fchannel.maxreq_sz;
3015 }
3016
3017 static bool replay_matches_cache(struct svc_rqst *rqstp,
3018                  struct nfsd4_sequence *seq, struct nfsd4_slot *slot)
3019 {
3020         struct nfsd4_compoundargs *argp = rqstp->rq_argp;
3021
3022         if ((bool)(slot->sl_flags & NFSD4_SLOT_CACHETHIS) !=
3023             (bool)seq->cachethis)
3024                 return false;
3025         /*
3026          * If there's an error than the reply can have fewer ops than
3027          * the call.  But if we cached a reply with *more* ops than the
3028          * call you're sending us now, then this new call is clearly not
3029          * really a replay of the old one:
3030          */
3031         if (slot->sl_opcnt < argp->opcnt)
3032                 return false;
3033         /* This is the only check explicitly called by spec: */
3034         if (!same_creds(&rqstp->rq_cred, &slot->sl_cred))
3035                 return false;
3036         /*
3037          * There may be more comparisons we could actually do, but the
3038          * spec doesn't require us to catch every case where the calls
3039          * don't match (that would require caching the call as well as
3040          * the reply), so we don't bother.
3041          */
3042         return true;
3043 }
3044
3045 __be32
3046 nfsd4_sequence(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3047                 union nfsd4_op_u *u)
3048 {
3049         struct nfsd4_sequence *seq = &u->sequence;
3050         struct nfsd4_compoundres *resp = rqstp->rq_resp;
3051         struct xdr_stream *xdr = &resp->xdr;
3052         struct nfsd4_session *session;
3053         struct nfs4_client *clp;
3054         struct nfsd4_slot *slot;
3055         struct nfsd4_conn *conn;
3056         __be32 status;
3057         int buflen;
3058         struct net *net = SVC_NET(rqstp);
3059         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
3060
3061         if (resp->opcnt != 1)
3062                 return nfserr_sequence_pos;
3063
3064         /*
3065          * Will be either used or freed by nfsd4_sequence_check_conn
3066          * below.
3067          */
3068         conn = alloc_conn(rqstp, NFS4_CDFC4_FORE);
3069         if (!conn)
3070                 return nfserr_jukebox;
3071
3072         spin_lock(&nn->client_lock);
3073         session = find_in_sessionid_hashtbl(&seq->sessionid, net, &status);
3074         if (!session)
3075                 goto out_no_session;
3076         clp = session->se_client;
3077
3078         status = nfserr_too_many_ops;
3079         if (nfsd4_session_too_many_ops(rqstp, session))
3080                 goto out_put_session;
3081
3082         status = nfserr_req_too_big;
3083         if (nfsd4_request_too_big(rqstp, session))
3084                 goto out_put_session;
3085
3086         status = nfserr_badslot;
3087         if (seq->slotid >= session->se_fchannel.maxreqs)
3088                 goto out_put_session;
3089
3090         slot = session->se_slots[seq->slotid];
3091         dprintk("%s: slotid %d\n", __func__, seq->slotid);
3092
3093         /* We do not negotiate the number of slots yet, so set the
3094          * maxslots to the session maxreqs which is used to encode
3095          * sr_highest_slotid and the sr_target_slot id to maxslots */
3096         seq->maxslots = session->se_fchannel.maxreqs;
3097
3098         status = check_slot_seqid(seq->seqid, slot->sl_seqid,
3099                                         slot->sl_flags & NFSD4_SLOT_INUSE);
3100         if (status == nfserr_replay_cache) {
3101                 status = nfserr_seq_misordered;
3102                 if (!(slot->sl_flags & NFSD4_SLOT_INITIALIZED))
3103                         goto out_put_session;
3104                 status = nfserr_seq_false_retry;
3105                 if (!replay_matches_cache(rqstp, seq, slot))
3106                         goto out_put_session;
3107                 cstate->slot = slot;
3108                 cstate->session = session;
3109                 cstate->clp = clp;
3110                 /* Return the cached reply status and set cstate->status
3111                  * for nfsd4_proc_compound processing */
3112                 status = nfsd4_replay_cache_entry(resp, seq);
3113                 cstate->status = nfserr_replay_cache;
3114                 goto out;
3115         }
3116         if (status)
3117                 goto out_put_session;
3118
3119         status = nfsd4_sequence_check_conn(conn, session);
3120         conn = NULL;
3121         if (status)
3122                 goto out_put_session;
3123
3124         buflen = (seq->cachethis) ?
3125                         session->se_fchannel.maxresp_cached :
3126                         session->se_fchannel.maxresp_sz;
3127         status = (seq->cachethis) ? nfserr_rep_too_big_to_cache :
3128                                     nfserr_rep_too_big;
3129         if (xdr_restrict_buflen(xdr, buflen - rqstp->rq_auth_slack))
3130                 goto out_put_session;
3131         svc_reserve(rqstp, buflen);
3132
3133         status = nfs_ok;
3134         /* Success! bump slot seqid */
3135         slot->sl_seqid = seq->seqid;
3136         slot->sl_flags |= NFSD4_SLOT_INUSE;
3137         if (seq->cachethis)
3138                 slot->sl_flags |= NFSD4_SLOT_CACHETHIS;
3139         else
3140                 slot->sl_flags &= ~NFSD4_SLOT_CACHETHIS;
3141
3142         cstate->slot = slot;
3143         cstate->session = session;
3144         cstate->clp = clp;
3145
3146 out:
3147         switch (clp->cl_cb_state) {
3148         case NFSD4_CB_DOWN:
3149                 seq->status_flags = SEQ4_STATUS_CB_PATH_DOWN;
3150                 break;
3151         case NFSD4_CB_FAULT:
3152                 seq->status_flags = SEQ4_STATUS_BACKCHANNEL_FAULT;
3153                 break;
3154         default:
3155                 seq->status_flags = 0;
3156         }
3157         if (!list_empty(&clp->cl_revoked))
3158                 seq->status_flags |= SEQ4_STATUS_RECALLABLE_STATE_REVOKED;
3159 out_no_session:
3160         if (conn)
3161                 free_conn(conn);
3162         spin_unlock(&nn->client_lock);
3163         return status;
3164 out_put_session:
3165         nfsd4_put_session_locked(session);
3166         goto out_no_session;
3167 }
3168
3169 void
3170 nfsd4_sequence_done(struct nfsd4_compoundres *resp)
3171 {
3172         struct nfsd4_compound_state *cs = &resp->cstate;
3173
3174         if (nfsd4_has_session(cs)) {
3175                 if (cs->status != nfserr_replay_cache) {
3176                         nfsd4_store_cache_entry(resp);
3177                         cs->slot->sl_flags &= ~NFSD4_SLOT_INUSE;
3178                 }
3179                 /* Drop session reference that was taken in nfsd4_sequence() */
3180                 nfsd4_put_session(cs->session);
3181         } else if (cs->clp)
3182                 put_client_renew(cs->clp);
3183 }
3184
3185 __be32
3186 nfsd4_destroy_clientid(struct svc_rqst *rqstp,
3187                 struct nfsd4_compound_state *cstate,
3188                 union nfsd4_op_u *u)
3189 {
3190         struct nfsd4_destroy_clientid *dc = &u->destroy_clientid;
3191         struct nfs4_client *conf, *unconf;
3192         struct nfs4_client *clp = NULL;
3193         __be32 status = 0;
3194         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3195
3196         spin_lock(&nn->client_lock);
3197         unconf = find_unconfirmed_client(&dc->clientid, true, nn);
3198         conf = find_confirmed_client(&dc->clientid, true, nn);
3199         WARN_ON_ONCE(conf && unconf);
3200
3201         if (conf) {
3202                 if (client_has_state(conf)) {
3203                         status = nfserr_clientid_busy;
3204                         goto out;
3205                 }
3206                 status = mark_client_expired_locked(conf);
3207                 if (status)
3208                         goto out;
3209                 clp = conf;
3210         } else if (unconf)
3211                 clp = unconf;
3212         else {
3213                 status = nfserr_stale_clientid;
3214                 goto out;
3215         }
3216         if (!nfsd4_mach_creds_match(clp, rqstp)) {
3217                 clp = NULL;
3218                 status = nfserr_wrong_cred;
3219                 goto out;
3220         }
3221         unhash_client_locked(clp);
3222 out:
3223         spin_unlock(&nn->client_lock);
3224         if (clp)
3225                 expire_client(clp);
3226         return status;
3227 }
3228
3229 __be32
3230 nfsd4_reclaim_complete(struct svc_rqst *rqstp,
3231                 struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
3232 {
3233         struct nfsd4_reclaim_complete *rc = &u->reclaim_complete;
3234         __be32 status = 0;
3235
3236         if (rc->rca_one_fs) {
3237                 if (!cstate->current_fh.fh_dentry)
3238                         return nfserr_nofilehandle;
3239                 /*
3240                  * We don't take advantage of the rca_one_fs case.
3241                  * That's OK, it's optional, we can safely ignore it.
3242                  */
3243                 return nfs_ok;
3244         }
3245
3246         status = nfserr_complete_already;
3247         if (test_and_set_bit(NFSD4_CLIENT_RECLAIM_COMPLETE,
3248                              &cstate->session->se_client->cl_flags))
3249                 goto out;
3250
3251         status = nfserr_stale_clientid;
3252         if (is_client_expired(cstate->session->se_client))
3253                 /*
3254                  * The following error isn't really legal.
3255                  * But we only get here if the client just explicitly
3256                  * destroyed the client.  Surely it no longer cares what
3257                  * error it gets back on an operation for the dead
3258                  * client.
3259                  */
3260                 goto out;
3261
3262         status = nfs_ok;
3263         nfsd4_client_record_create(cstate->session->se_client);
3264 out:
3265         return status;
3266 }
3267
3268 __be32
3269 nfsd4_setclientid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
3270                   union nfsd4_op_u *u)
3271 {
3272         struct nfsd4_setclientid *setclid = &u->setclientid;
3273         struct xdr_netobj       clname = setclid->se_name;
3274         nfs4_verifier           clverifier = setclid->se_verf;
3275         struct nfs4_client      *conf, *new;
3276         struct nfs4_client      *unconf = NULL;
3277         __be32                  status;
3278         struct nfsd_net         *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3279
3280         new = create_client(clname, rqstp, &clverifier);
3281         if (new == NULL)
3282                 return nfserr_jukebox;
3283         /* Cases below refer to rfc 3530 section 14.2.33: */
3284         spin_lock(&nn->client_lock);
3285         conf = find_confirmed_client_by_name(&clname, nn);
3286         if (conf && client_has_state(conf)) {
3287                 /* case 0: */
3288                 status = nfserr_clid_inuse;
3289                 if (clp_used_exchangeid(conf))
3290                         goto out;
3291                 if (!same_creds(&conf->cl_cred, &rqstp->rq_cred)) {
3292                         char addr_str[INET6_ADDRSTRLEN];
3293                         rpc_ntop((struct sockaddr *) &conf->cl_addr, addr_str,
3294                                  sizeof(addr_str));
3295                         dprintk("NFSD: setclientid: string in use by client "
3296                                 "at %s\n", addr_str);
3297                         goto out;
3298                 }
3299         }
3300         unconf = find_unconfirmed_client_by_name(&clname, nn);
3301         if (unconf)
3302                 unhash_client_locked(unconf);
3303         if (conf && same_verf(&conf->cl_verifier, &clverifier)) {
3304                 /* case 1: probable callback update */
3305                 copy_clid(new, conf);
3306                 gen_confirm(new, nn);
3307         } else /* case 4 (new client) or cases 2, 3 (client reboot): */
3308                 gen_clid(new, nn);
3309         new->cl_minorversion = 0;
3310         gen_callback(new, setclid, rqstp);
3311         add_to_unconfirmed(new);
3312         setclid->se_clientid.cl_boot = new->cl_clientid.cl_boot;
3313         setclid->se_clientid.cl_id = new->cl_clientid.cl_id;
3314         memcpy(setclid->se_confirm.data, new->cl_confirm.data, sizeof(setclid->se_confirm.data));
3315         new = NULL;
3316         status = nfs_ok;
3317 out:
3318         spin_unlock(&nn->client_lock);
3319         if (new)
3320                 free_client(new);
3321         if (unconf)
3322                 expire_client(unconf);
3323         return status;
3324 }
3325
3326
3327 __be32
3328 nfsd4_setclientid_confirm(struct svc_rqst *rqstp,
3329                         struct nfsd4_compound_state *cstate,
3330                         union nfsd4_op_u *u)
3331 {
3332         struct nfsd4_setclientid_confirm *setclientid_confirm =
3333                         &u->setclientid_confirm;
3334         struct nfs4_client *conf, *unconf;
3335         struct nfs4_client *old = NULL;
3336         nfs4_verifier confirm = setclientid_confirm->sc_confirm;
3337         clientid_t * clid = &setclientid_confirm->sc_clientid;
3338         __be32 status;
3339         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
3340
3341         if (STALE_CLIENTID(clid, nn))
3342                 return nfserr_stale_clientid;
3343
3344         spin_lock(&nn->client_lock);
3345         conf = find_confirmed_client(clid, false, nn);
3346         unconf = find_unconfirmed_client(clid, false, nn);
3347         /*
3348          * We try hard to give out unique clientid's, so if we get an
3349          * attempt to confirm the same clientid with a different cred,
3350          * the client may be buggy; this should never happen.
3351          *
3352          * Nevertheless, RFC 7530 recommends INUSE for this case:
3353          */
3354         status = nfserr_clid_inuse;
3355         if (unconf && !same_creds(&unconf->cl_cred, &rqstp->rq_cred))
3356                 goto out;
3357         if (conf && !same_creds(&conf->cl_cred, &rqstp->rq_cred))
3358                 goto out;
3359         /* cases below refer to rfc 3530 section 14.2.34: */
3360         if (!unconf || !same_verf(&confirm, &unconf->cl_confirm)) {
3361                 if (conf && same_verf(&confirm, &conf->cl_confirm)) {
3362                         /* case 2: probable retransmit */
3363                         status = nfs_ok;
3364                 } else /* case 4: client hasn't noticed we rebooted yet? */
3365                         status = nfserr_stale_clientid;
3366                 goto out;
3367         }
3368         status = nfs_ok;
3369         if (conf) { /* case 1: callback update */
3370                 old = unconf;
3371                 unhash_client_locked(old);
3372                 nfsd4_change_callback(conf, &unconf->cl_cb_conn);
3373         } else { /* case 3: normal case; new or rebooted client */
3374                 old = find_confirmed_client_by_name(&unconf->cl_name, nn);
3375                 if (old) {
3376                         status = nfserr_clid_inuse;
3377                         if (client_has_state(old)
3378                                         && !same_creds(&unconf->cl_cred,
3379                                                         &old->cl_cred))
3380                                 goto out;
3381                         status = mark_client_expired_locked(old);
3382                         if (status) {
3383                                 old = NULL;
3384                                 goto out;
3385                         }
3386                 }
3387                 move_to_confirmed(unconf);
3388                 conf = unconf;
3389         }
3390         get_client_locked(conf);
3391         spin_unlock(&nn->client_lock);
3392         nfsd4_probe_callback(conf);
3393         spin_lock(&nn->client_lock);
3394         put_client_renew_locked(conf);
3395 out:
3396         spin_unlock(&nn->client_lock);
3397         if (old)
3398                 expire_client(old);
3399         return status;
3400 }
3401
3402 static struct nfs4_file *nfsd4_alloc_file(void)
3403 {
3404         return kmem_cache_alloc(file_slab, GFP_KERNEL);
3405 }
3406
3407 /* OPEN Share state helper functions */
3408 static void nfsd4_init_file(struct knfsd_fh *fh, unsigned int hashval,
3409                                 struct nfs4_file *fp)
3410 {
3411         lockdep_assert_held(&state_lock);
3412
3413         refcount_set(&fp->fi_ref, 1);
3414         spin_lock_init(&fp->fi_lock);
3415         INIT_LIST_HEAD(&fp->fi_stateids);
3416         INIT_LIST_HEAD(&fp->fi_delegations);
3417         INIT_LIST_HEAD(&fp->fi_clnt_odstate);
3418         fh_copy_shallow(&fp->fi_fhandle, fh);
3419         fp->fi_deleg_file = NULL;
3420         fp->fi_had_conflict = false;
3421         fp->fi_share_deny = 0;
3422         memset(fp->fi_fds, 0, sizeof(fp->fi_fds));
3423         memset(fp->fi_access, 0, sizeof(fp->fi_access));
3424 #ifdef CONFIG_NFSD_PNFS
3425         INIT_LIST_HEAD(&fp->fi_lo_states);
3426         atomic_set(&fp->fi_lo_recalls, 0);
3427 #endif
3428         hlist_add_head_rcu(&fp->fi_hash, &file_hashtbl[hashval]);
3429 }
3430
3431 void
3432 nfsd4_free_slabs(void)
3433 {
3434         kmem_cache_destroy(odstate_slab);
3435         kmem_cache_destroy(openowner_slab);
3436         kmem_cache_destroy(lockowner_slab);
3437         kmem_cache_destroy(file_slab);
3438         kmem_cache_destroy(stateid_slab);
3439         kmem_cache_destroy(deleg_slab);
3440 }
3441
3442 int
3443 nfsd4_init_slabs(void)
3444 {
3445         openowner_slab = kmem_cache_create("nfsd4_openowners",
3446                         sizeof(struct nfs4_openowner), 0, 0, NULL);
3447         if (openowner_slab == NULL)
3448                 goto out;
3449         lockowner_slab = kmem_cache_create("nfsd4_lockowners",
3450                         sizeof(struct nfs4_lockowner), 0, 0, NULL);
3451         if (lockowner_slab == NULL)
3452                 goto out_free_openowner_slab;
3453         file_slab = kmem_cache_create("nfsd4_files",
3454                         sizeof(struct nfs4_file), 0, 0, NULL);
3455         if (file_slab == NULL)
3456                 goto out_free_lockowner_slab;
3457         stateid_slab = kmem_cache_create("nfsd4_stateids",
3458                         sizeof(struct nfs4_ol_stateid), 0, 0, NULL);
3459         if (stateid_slab == NULL)
3460                 goto out_free_file_slab;
3461         deleg_slab = kmem_cache_create("nfsd4_delegations",
3462                         sizeof(struct nfs4_delegation), 0, 0, NULL);
3463         if (deleg_slab == NULL)
3464                 goto out_free_stateid_slab;
3465         odstate_slab = kmem_cache_create("nfsd4_odstate",
3466                         sizeof(struct nfs4_clnt_odstate), 0, 0, NULL);
3467         if (odstate_slab == NULL)
3468                 goto out_free_deleg_slab;
3469         return 0;
3470
3471 out_free_deleg_slab:
3472         kmem_cache_destroy(deleg_slab);
3473 out_free_stateid_slab:
3474         kmem_cache_destroy(stateid_slab);
3475 out_free_file_slab:
3476         kmem_cache_destroy(file_slab);
3477 out_free_lockowner_slab:
3478         kmem_cache_destroy(lockowner_slab);
3479 out_free_openowner_slab:
3480         kmem_cache_destroy(openowner_slab);
3481 out:
3482         dprintk("nfsd4: out of memory while initializing nfsv4\n");
3483         return -ENOMEM;
3484 }
3485
3486 static void init_nfs4_replay(struct nfs4_replay *rp)
3487 {
3488         rp->rp_status = nfserr_serverfault;
3489         rp->rp_buflen = 0;
3490         rp->rp_buf = rp->rp_ibuf;
3491         mutex_init(&rp->rp_mutex);
3492 }
3493
3494 static void nfsd4_cstate_assign_replay(struct nfsd4_compound_state *cstate,
3495                 struct nfs4_stateowner *so)
3496 {
3497         if (!nfsd4_has_session(cstate)) {
3498                 mutex_lock(&so->so_replay.rp_mutex);
3499                 cstate->replay_owner = nfs4_get_stateowner(so);
3500         }
3501 }
3502
3503 void nfsd4_cstate_clear_replay(struct nfsd4_compound_state *cstate)
3504 {
3505         struct nfs4_stateowner *so = cstate->replay_owner;
3506
3507         if (so != NULL) {
3508                 cstate->replay_owner = NULL;
3509                 mutex_unlock(&so->so_replay.rp_mutex);
3510                 nfs4_put_stateowner(so);
3511         }
3512 }
3513
3514 static inline void *alloc_stateowner(struct kmem_cache *slab, struct xdr_netobj *owner, struct nfs4_client *clp)
3515 {
3516         struct nfs4_stateowner *sop;
3517
3518         sop = kmem_cache_alloc(slab, GFP_KERNEL);
3519         if (!sop)
3520                 return NULL;
3521
3522         sop->so_owner.data = kmemdup(owner->data, owner->len, GFP_KERNEL);
3523         if (!sop->so_owner.data) {
3524                 kmem_cache_free(slab, sop);
3525                 return NULL;
3526         }
3527         sop->so_owner.len = owner->len;
3528
3529         INIT_LIST_HEAD(&sop->so_stateids);
3530         sop->so_client = clp;
3531         init_nfs4_replay(&sop->so_replay);
3532         atomic_set(&sop->so_count, 1);
3533         return sop;
3534 }
3535
3536 static void hash_openowner(struct nfs4_openowner *oo, struct nfs4_client *clp, unsigned int strhashval)
3537 {
3538         lockdep_assert_held(&clp->cl_lock);
3539
3540         list_add(&oo->oo_owner.so_strhash,
3541                  &clp->cl_ownerstr_hashtbl[strhashval]);
3542         list_add(&oo->oo_perclient, &clp->cl_openowners);
3543 }
3544
3545 static void nfs4_unhash_openowner(struct nfs4_stateowner *so)
3546 {
3547         unhash_openowner_locked(openowner(so));
3548 }
3549
3550 static void nfs4_free_openowner(struct nfs4_stateowner *so)
3551 {
3552         struct nfs4_openowner *oo = openowner(so);
3553
3554         kmem_cache_free(openowner_slab, oo);
3555 }
3556
3557 static const struct nfs4_stateowner_operations openowner_ops = {
3558         .so_unhash =    nfs4_unhash_openowner,
3559         .so_free =      nfs4_free_openowner,
3560 };
3561
3562 static struct nfs4_ol_stateid *
3563 nfsd4_find_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3564 {
3565         struct nfs4_ol_stateid *local, *ret = NULL;
3566         struct nfs4_openowner *oo = open->op_openowner;
3567
3568         lockdep_assert_held(&fp->fi_lock);
3569
3570         list_for_each_entry(local, &fp->fi_stateids, st_perfile) {
3571                 /* ignore lock owners */
3572                 if (local->st_stateowner->so_is_open_owner == 0)
3573                         continue;
3574                 if (local->st_stateowner != &oo->oo_owner)
3575                         continue;
3576                 if (local->st_stid.sc_type == NFS4_OPEN_STID) {
3577                         ret = local;
3578                         refcount_inc(&ret->st_stid.sc_count);
3579                         break;
3580                 }
3581         }
3582         return ret;
3583 }
3584
3585 static __be32
3586 nfsd4_verify_open_stid(struct nfs4_stid *s)
3587 {
3588         __be32 ret = nfs_ok;
3589
3590         switch (s->sc_type) {
3591         default:
3592                 break;
3593         case NFS4_CLOSED_STID:
3594         case NFS4_CLOSED_DELEG_STID:
3595                 ret = nfserr_bad_stateid;
3596                 break;
3597         case NFS4_REVOKED_DELEG_STID:
3598                 ret = nfserr_deleg_revoked;
3599         }
3600         return ret;
3601 }
3602
3603 /* Lock the stateid st_mutex, and deal with races with CLOSE */
3604 static __be32
3605 nfsd4_lock_ol_stateid(struct nfs4_ol_stateid *stp)
3606 {
3607         __be32 ret;
3608
3609         mutex_lock_nested(&stp->st_mutex, LOCK_STATEID_MUTEX);
3610         ret = nfsd4_verify_open_stid(&stp->st_stid);
3611         if (ret != nfs_ok)
3612                 mutex_unlock(&stp->st_mutex);
3613         return ret;
3614 }
3615
3616 static struct nfs4_ol_stateid *
3617 nfsd4_find_and_lock_existing_open(struct nfs4_file *fp, struct nfsd4_open *open)
3618 {
3619         struct nfs4_ol_stateid *stp;
3620         for (;;) {
3621                 spin_lock(&fp->fi_lock);
3622                 stp = nfsd4_find_existing_open(fp, open);
3623                 spin_unlock(&fp->fi_lock);
3624                 if (!stp || nfsd4_lock_ol_stateid(stp) == nfs_ok)
3625                         break;
3626                 nfs4_put_stid(&stp->st_stid);
3627         }
3628         return stp;
3629 }
3630
3631 static struct nfs4_openowner *
3632 alloc_init_open_stateowner(unsigned int strhashval, struct nfsd4_open *open,
3633                            struct nfsd4_compound_state *cstate)
3634 {
3635         struct nfs4_client *clp = cstate->clp;
3636         struct nfs4_openowner *oo, *ret;
3637
3638         oo = alloc_stateowner(openowner_slab, &open->op_owner, clp);
3639         if (!oo)
3640                 return NULL;
3641         oo->oo_owner.so_ops = &openowner_ops;
3642         oo->oo_owner.so_is_open_owner = 1;
3643         oo->oo_owner.so_seqid = open->op_seqid;
3644         oo->oo_flags = 0;
3645         if (nfsd4_has_session(cstate))
3646                 oo->oo_flags |= NFS4_OO_CONFIRMED;
3647         oo->oo_time = 0;
3648         oo->oo_last_closed_stid = NULL;
3649         INIT_LIST_HEAD(&oo->oo_close_lru);
3650         spin_lock(&clp->cl_lock);
3651         ret = find_openstateowner_str_locked(strhashval, open, clp);
3652         if (ret == NULL) {
3653                 hash_openowner(oo, clp, strhashval);
3654                 ret = oo;
3655         } else
3656                 nfs4_free_stateowner(&oo->oo_owner);
3657
3658         spin_unlock(&clp->cl_lock);
3659         return ret;
3660 }
3661
3662 static struct nfs4_ol_stateid *
3663 init_open_stateid(struct nfs4_file *fp, struct nfsd4_open *open)
3664 {
3665
3666         struct nfs4_openowner *oo = open->op_openowner;
3667         struct nfs4_ol_stateid *retstp = NULL;
3668         struct nfs4_ol_stateid *stp;
3669
3670         stp = open->op_stp;
3671         /* We are moving these outside of the spinlocks to avoid the warnings */
3672         mutex_init(&stp->st_mutex);
3673         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
3674
3675 retry:
3676         spin_lock(&oo->oo_owner.so_client->cl_lock);
3677         spin_lock(&fp->fi_lock);
3678
3679         retstp = nfsd4_find_existing_open(fp, open);
3680         if (retstp)
3681                 goto out_unlock;
3682
3683         open->op_stp = NULL;
3684         refcount_inc(&stp->st_stid.sc_count);
3685         stp->st_stid.sc_type = NFS4_OPEN_STID;
3686         INIT_LIST_HEAD(&stp->st_locks);
3687         stp->st_stateowner = nfs4_get_stateowner(&oo->oo_owner);
3688         get_nfs4_file(fp);
3689         stp->st_stid.sc_file = fp;
3690         stp->st_access_bmap = 0;
3691         stp->st_deny_bmap = 0;
3692         stp->st_openstp = NULL;
3693         list_add(&stp->st_perstateowner, &oo->oo_owner.so_stateids);
3694         list_add(&stp->st_perfile, &fp->fi_stateids);
3695
3696 out_unlock:
3697         spin_unlock(&fp->fi_lock);
3698         spin_unlock(&oo->oo_owner.so_client->cl_lock);
3699         if (retstp) {
3700                 /* Handle races with CLOSE */
3701                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
3702                         nfs4_put_stid(&retstp->st_stid);
3703                         goto retry;
3704                 }
3705                 /* To keep mutex tracking happy */
3706                 mutex_unlock(&stp->st_mutex);
3707                 stp = retstp;
3708         }
3709         return stp;
3710 }
3711
3712 /*
3713  * In the 4.0 case we need to keep the owners around a little while to handle
3714  * CLOSE replay. We still do need to release any file access that is held by
3715  * them before returning however.
3716  */
3717 static void
3718 move_to_close_lru(struct nfs4_ol_stateid *s, struct net *net)
3719 {
3720         struct nfs4_ol_stateid *last;
3721         struct nfs4_openowner *oo = openowner(s->st_stateowner);
3722         struct nfsd_net *nn = net_generic(s->st_stid.sc_client->net,
3723                                                 nfsd_net_id);
3724
3725         dprintk("NFSD: move_to_close_lru nfs4_openowner %p\n", oo);
3726
3727         /*
3728          * We know that we hold one reference via nfsd4_close, and another
3729          * "persistent" reference for the client. If the refcount is higher
3730          * than 2, then there are still calls in progress that are using this
3731          * stateid. We can't put the sc_file reference until they are finished.
3732          * Wait for the refcount to drop to 2. Since it has been unhashed,
3733          * there should be no danger of the refcount going back up again at
3734          * this point.
3735          */
3736         wait_event(close_wq, refcount_read(&s->st_stid.sc_count) == 2);
3737
3738         release_all_access(s);
3739         if (s->st_stid.sc_file) {
3740                 put_nfs4_file(s->st_stid.sc_file);
3741                 s->st_stid.sc_file = NULL;
3742         }
3743
3744         spin_lock(&nn->client_lock);
3745         last = oo->oo_last_closed_stid;
3746         oo->oo_last_closed_stid = s;
3747         list_move_tail(&oo->oo_close_lru, &nn->close_lru);
3748         oo->oo_time = get_seconds();
3749         spin_unlock(&nn->client_lock);
3750         if (last)
3751                 nfs4_put_stid(&last->st_stid);
3752 }
3753
3754 /* search file_hashtbl[] for file */
3755 static struct nfs4_file *
3756 find_file_locked(struct knfsd_fh *fh, unsigned int hashval)
3757 {
3758         struct nfs4_file *fp;
3759
3760         hlist_for_each_entry_rcu(fp, &file_hashtbl[hashval], fi_hash) {
3761                 if (fh_match(&fp->fi_fhandle, fh)) {
3762                         if (refcount_inc_not_zero(&fp->fi_ref))
3763                                 return fp;
3764                 }
3765         }
3766         return NULL;
3767 }
3768
3769 struct nfs4_file *
3770 find_file(struct knfsd_fh *fh)
3771 {
3772         struct nfs4_file *fp;
3773         unsigned int hashval = file_hashval(fh);
3774
3775         rcu_read_lock();
3776         fp = find_file_locked(fh, hashval);
3777         rcu_read_unlock();
3778         return fp;
3779 }
3780
3781 static struct nfs4_file *
3782 find_or_add_file(struct nfs4_file *new, struct knfsd_fh *fh)
3783 {
3784         struct nfs4_file *fp;
3785         unsigned int hashval = file_hashval(fh);
3786
3787         rcu_read_lock();
3788         fp = find_file_locked(fh, hashval);
3789         rcu_read_unlock();
3790         if (fp)
3791                 return fp;
3792
3793         spin_lock(&state_lock);
3794         fp = find_file_locked(fh, hashval);
3795         if (likely(fp == NULL)) {
3796                 nfsd4_init_file(fh, hashval, new);
3797                 fp = new;
3798         }
3799         spin_unlock(&state_lock);
3800
3801         return fp;
3802 }
3803
3804 /*
3805  * Called to check deny when READ with all zero stateid or
3806  * WRITE with all zero or all one stateid
3807  */
3808 static __be32
3809 nfs4_share_conflict(struct svc_fh *current_fh, unsigned int deny_type)
3810 {
3811         struct nfs4_file *fp;
3812         __be32 ret = nfs_ok;
3813
3814         fp = find_file(&current_fh->fh_handle);
3815         if (!fp)
3816                 return ret;
3817         /* Check for conflicting share reservations */
3818         spin_lock(&fp->fi_lock);
3819         if (fp->fi_share_deny & deny_type)
3820                 ret = nfserr_locked;
3821         spin_unlock(&fp->fi_lock);
3822         put_nfs4_file(fp);
3823         return ret;
3824 }
3825
3826 static void nfsd4_cb_recall_prepare(struct nfsd4_callback *cb)
3827 {
3828         struct nfs4_delegation *dp = cb_to_delegation(cb);
3829         struct nfsd_net *nn = net_generic(dp->dl_stid.sc_client->net,
3830                                           nfsd_net_id);
3831
3832         block_delegations(&dp->dl_stid.sc_file->fi_fhandle);
3833
3834         /*
3835          * We can't do this in nfsd_break_deleg_cb because it is
3836          * already holding inode->i_lock.
3837          *
3838          * If the dl_time != 0, then we know that it has already been
3839          * queued for a lease break. Don't queue it again.
3840          */
3841         spin_lock(&state_lock);
3842         if (dp->dl_time == 0) {
3843                 dp->dl_time = get_seconds();
3844                 list_add_tail(&dp->dl_recall_lru, &nn->del_recall_lru);
3845         }
3846         spin_unlock(&state_lock);
3847 }
3848
3849 static int nfsd4_cb_recall_done(struct nfsd4_callback *cb,
3850                 struct rpc_task *task)
3851 {
3852         struct nfs4_delegation *dp = cb_to_delegation(cb);
3853
3854         if (dp->dl_stid.sc_type == NFS4_CLOSED_DELEG_STID)
3855                 return 1;
3856
3857         switch (task->tk_status) {
3858         case 0:
3859                 return 1;
3860         case -EBADHANDLE:
3861         case -NFS4ERR_BAD_STATEID:
3862                 /*
3863                  * Race: client probably got cb_recall before open reply
3864                  * granting delegation.
3865                  */
3866                 if (dp->dl_retries--) {
3867                         rpc_delay(task, 2 * HZ);
3868                         return 0;
3869                 }
3870                 /*FALLTHRU*/
3871         default:
3872                 return -1;
3873         }
3874 }
3875
3876 static void nfsd4_cb_recall_release(struct nfsd4_callback *cb)
3877 {
3878         struct nfs4_delegation *dp = cb_to_delegation(cb);
3879
3880         nfs4_put_stid(&dp->dl_stid);
3881 }
3882
3883 static const struct nfsd4_callback_ops nfsd4_cb_recall_ops = {
3884         .prepare        = nfsd4_cb_recall_prepare,
3885         .done           = nfsd4_cb_recall_done,
3886         .release        = nfsd4_cb_recall_release,
3887 };
3888
3889 static void nfsd_break_one_deleg(struct nfs4_delegation *dp)
3890 {
3891         /*
3892          * We're assuming the state code never drops its reference
3893          * without first removing the lease.  Since we're in this lease
3894          * callback (and since the lease code is serialized by the kernel
3895          * lock) we know the server hasn't removed the lease yet, we know
3896          * it's safe to take a reference.
3897          */
3898         refcount_inc(&dp->dl_stid.sc_count);
3899         nfsd4_run_cb(&dp->dl_recall);
3900 }
3901
3902 /* Called from break_lease() with i_lock held. */
3903 static bool
3904 nfsd_break_deleg_cb(struct file_lock *fl)
3905 {
3906         bool ret = false;
3907         struct nfs4_file *fp = (struct nfs4_file *)fl->fl_owner;
3908         struct nfs4_delegation *dp;
3909
3910         if (!fp) {
3911                 WARN(1, "(%p)->fl_owner NULL\n", fl);
3912                 return ret;
3913         }
3914         if (fp->fi_had_conflict) {
3915                 WARN(1, "duplicate break on %p\n", fp);
3916                 return ret;
3917         }
3918         /*
3919          * We don't want the locks code to timeout the lease for us;
3920          * we'll remove it ourself if a delegation isn't returned
3921          * in time:
3922          */
3923         fl->fl_break_time = 0;
3924
3925         spin_lock(&fp->fi_lock);
3926         fp->fi_had_conflict = true;
3927         /*
3928          * If there are no delegations on the list, then return true
3929          * so that the lease code will go ahead and delete it.
3930          */
3931         if (list_empty(&fp->fi_delegations))
3932                 ret = true;
3933         else
3934                 list_for_each_entry(dp, &fp->fi_delegations, dl_perfile)
3935                         nfsd_break_one_deleg(dp);
3936         spin_unlock(&fp->fi_lock);
3937         return ret;
3938 }
3939
3940 static int
3941 nfsd_change_deleg_cb(struct file_lock *onlist, int arg,
3942                      struct list_head *dispose)
3943 {
3944         if (arg & F_UNLCK)
3945                 return lease_modify(onlist, arg, dispose);
3946         else
3947                 return -EAGAIN;
3948 }
3949
3950 static const struct lock_manager_operations nfsd_lease_mng_ops = {
3951         .lm_break = nfsd_break_deleg_cb,
3952         .lm_change = nfsd_change_deleg_cb,
3953 };
3954
3955 static __be32 nfsd4_check_seqid(struct nfsd4_compound_state *cstate, struct nfs4_stateowner *so, u32 seqid)
3956 {
3957         if (nfsd4_has_session(cstate))
3958                 return nfs_ok;
3959         if (seqid == so->so_seqid - 1)
3960                 return nfserr_replay_me;
3961         if (seqid == so->so_seqid)
3962                 return nfs_ok;
3963         return nfserr_bad_seqid;
3964 }
3965
3966 static __be32 lookup_clientid(clientid_t *clid,
3967                 struct nfsd4_compound_state *cstate,
3968                 struct nfsd_net *nn)
3969 {
3970         struct nfs4_client *found;
3971
3972         if (cstate->clp) {
3973                 found = cstate->clp;
3974                 if (!same_clid(&found->cl_clientid, clid))
3975                         return nfserr_stale_clientid;
3976                 return nfs_ok;
3977         }
3978
3979         if (STALE_CLIENTID(clid, nn))
3980                 return nfserr_stale_clientid;
3981
3982         /*
3983          * For v4.1+ we get the client in the SEQUENCE op. If we don't have one
3984          * cached already then we know this is for is for v4.0 and "sessions"
3985          * will be false.
3986          */
3987         WARN_ON_ONCE(cstate->session);
3988         spin_lock(&nn->client_lock);
3989         found = find_confirmed_client(clid, false, nn);
3990         if (!found) {
3991                 spin_unlock(&nn->client_lock);
3992                 return nfserr_expired;
3993         }
3994         atomic_inc(&found->cl_refcount);
3995         spin_unlock(&nn->client_lock);
3996
3997         /* Cache the nfs4_client in cstate! */
3998         cstate->clp = found;
3999         return nfs_ok;
4000 }
4001
4002 __be32
4003 nfsd4_process_open1(struct nfsd4_compound_state *cstate,
4004                     struct nfsd4_open *open, struct nfsd_net *nn)
4005 {
4006         clientid_t *clientid = &open->op_clientid;
4007         struct nfs4_client *clp = NULL;
4008         unsigned int strhashval;
4009         struct nfs4_openowner *oo = NULL;
4010         __be32 status;
4011
4012         if (STALE_CLIENTID(&open->op_clientid, nn))
4013                 return nfserr_stale_clientid;
4014         /*
4015          * In case we need it later, after we've already created the
4016          * file and don't want to risk a further failure:
4017          */
4018         open->op_file = nfsd4_alloc_file();
4019         if (open->op_file == NULL)
4020                 return nfserr_jukebox;
4021
4022         status = lookup_clientid(clientid, cstate, nn);
4023         if (status)
4024                 return status;
4025         clp = cstate->clp;
4026
4027         strhashval = ownerstr_hashval(&open->op_owner);
4028         oo = find_openstateowner_str(strhashval, open, clp);
4029         open->op_openowner = oo;
4030         if (!oo) {
4031                 goto new_owner;
4032         }
4033         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
4034                 /* Replace unconfirmed owners without checking for replay. */
4035                 release_openowner(oo);
4036                 open->op_openowner = NULL;
4037                 goto new_owner;
4038         }
4039         status = nfsd4_check_seqid(cstate, &oo->oo_owner, open->op_seqid);
4040         if (status)
4041                 return status;
4042         goto alloc_stateid;
4043 new_owner:
4044         oo = alloc_init_open_stateowner(strhashval, open, cstate);
4045         if (oo == NULL)
4046                 return nfserr_jukebox;
4047         open->op_openowner = oo;
4048 alloc_stateid:
4049         open->op_stp = nfs4_alloc_open_stateid(clp);
4050         if (!open->op_stp)
4051                 return nfserr_jukebox;
4052
4053         if (nfsd4_has_session(cstate) &&
4054             (cstate->current_fh.fh_export->ex_flags & NFSEXP_PNFS)) {
4055                 open->op_odstate = alloc_clnt_odstate(clp);
4056                 if (!open->op_odstate)
4057                         return nfserr_jukebox;
4058         }
4059
4060         return nfs_ok;
4061 }
4062
4063 static inline __be32
4064 nfs4_check_delegmode(struct nfs4_delegation *dp, int flags)
4065 {
4066         if ((flags & WR_STATE) && (dp->dl_type == NFS4_OPEN_DELEGATE_READ))
4067                 return nfserr_openmode;
4068         else
4069                 return nfs_ok;
4070 }
4071
4072 static int share_access_to_flags(u32 share_access)
4073 {
4074         return share_access == NFS4_SHARE_ACCESS_READ ? RD_STATE : WR_STATE;
4075 }
4076
4077 static struct nfs4_delegation *find_deleg_stateid(struct nfs4_client *cl, stateid_t *s)
4078 {
4079         struct nfs4_stid *ret;
4080
4081         ret = find_stateid_by_type(cl, s,
4082                                 NFS4_DELEG_STID|NFS4_REVOKED_DELEG_STID);
4083         if (!ret)
4084                 return NULL;
4085         return delegstateid(ret);
4086 }
4087
4088 static bool nfsd4_is_deleg_cur(struct nfsd4_open *open)
4089 {
4090         return open->op_claim_type == NFS4_OPEN_CLAIM_DELEGATE_CUR ||
4091                open->op_claim_type == NFS4_OPEN_CLAIM_DELEG_CUR_FH;
4092 }
4093
4094 static __be32
4095 nfs4_check_deleg(struct nfs4_client *cl, struct nfsd4_open *open,
4096                 struct nfs4_delegation **dp)
4097 {
4098         int flags;
4099         __be32 status = nfserr_bad_stateid;
4100         struct nfs4_delegation *deleg;
4101
4102         deleg = find_deleg_stateid(cl, &open->op_delegate_stateid);
4103         if (deleg == NULL)
4104                 goto out;
4105         if (deleg->dl_stid.sc_type == NFS4_REVOKED_DELEG_STID) {
4106                 nfs4_put_stid(&deleg->dl_stid);
4107                 if (cl->cl_minorversion)
4108                         status = nfserr_deleg_revoked;
4109                 goto out;
4110         }
4111         flags = share_access_to_flags(open->op_share_access);
4112         status = nfs4_check_delegmode(deleg, flags);
4113         if (status) {
4114                 nfs4_put_stid(&deleg->dl_stid);
4115                 goto out;
4116         }
4117         *dp = deleg;
4118 out:
4119         if (!nfsd4_is_deleg_cur(open))
4120                 return nfs_ok;
4121         if (status)
4122                 return status;
4123         open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4124         return nfs_ok;
4125 }
4126
4127 static inline int nfs4_access_to_access(u32 nfs4_access)
4128 {
4129         int flags = 0;
4130
4131         if (nfs4_access & NFS4_SHARE_ACCESS_READ)
4132                 flags |= NFSD_MAY_READ;
4133         if (nfs4_access & NFS4_SHARE_ACCESS_WRITE)
4134                 flags |= NFSD_MAY_WRITE;
4135         return flags;
4136 }
4137
4138 static inline __be32
4139 nfsd4_truncate(struct svc_rqst *rqstp, struct svc_fh *fh,
4140                 struct nfsd4_open *open)
4141 {
4142         struct iattr iattr = {
4143                 .ia_valid = ATTR_SIZE,
4144                 .ia_size = 0,
4145         };
4146         if (!open->op_truncate)
4147                 return 0;
4148         if (!(open->op_share_access & NFS4_SHARE_ACCESS_WRITE))
4149                 return nfserr_inval;
4150         return nfsd_setattr(rqstp, fh, &iattr, 0, (time_t)0);
4151 }
4152
4153 static __be32 nfs4_get_vfs_file(struct svc_rqst *rqstp, struct nfs4_file *fp,
4154                 struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp,
4155                 struct nfsd4_open *open)
4156 {
4157         struct file *filp = NULL;
4158         __be32 status;
4159         int oflag = nfs4_access_to_omode(open->op_share_access);
4160         int access = nfs4_access_to_access(open->op_share_access);
4161         unsigned char old_access_bmap, old_deny_bmap;
4162
4163         spin_lock(&fp->fi_lock);
4164
4165         /*
4166          * Are we trying to set a deny mode that would conflict with
4167          * current access?
4168          */
4169         status = nfs4_file_check_deny(fp, open->op_share_deny);
4170         if (status != nfs_ok) {
4171                 spin_unlock(&fp->fi_lock);
4172                 goto out;
4173         }
4174
4175         /* set access to the file */
4176         status = nfs4_file_get_access(fp, open->op_share_access);
4177         if (status != nfs_ok) {
4178                 spin_unlock(&fp->fi_lock);
4179                 goto out;
4180         }
4181
4182         /* Set access bits in stateid */
4183         old_access_bmap = stp->st_access_bmap;
4184         set_access(open->op_share_access, stp);
4185
4186         /* Set new deny mask */
4187         old_deny_bmap = stp->st_deny_bmap;
4188         set_deny(open->op_share_deny, stp);
4189         fp->fi_share_deny |= (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4190
4191         if (!fp->fi_fds[oflag]) {
4192                 spin_unlock(&fp->fi_lock);
4193                 status = nfsd_open(rqstp, cur_fh, S_IFREG, access, &filp);
4194                 if (status)
4195                         goto out_put_access;
4196                 spin_lock(&fp->fi_lock);
4197                 if (!fp->fi_fds[oflag]) {
4198                         fp->fi_fds[oflag] = filp;
4199                         filp = NULL;
4200                 }
4201         }
4202         spin_unlock(&fp->fi_lock);
4203         if (filp)
4204                 fput(filp);
4205
4206         status = nfsd4_truncate(rqstp, cur_fh, open);
4207         if (status)
4208                 goto out_put_access;
4209 out:
4210         return status;
4211 out_put_access:
4212         stp->st_access_bmap = old_access_bmap;
4213         nfs4_file_put_access(fp, open->op_share_access);
4214         reset_union_bmap_deny(bmap_to_share_mode(old_deny_bmap), stp);
4215         goto out;
4216 }
4217
4218 static __be32
4219 nfs4_upgrade_open(struct svc_rqst *rqstp, struct nfs4_file *fp, struct svc_fh *cur_fh, struct nfs4_ol_stateid *stp, struct nfsd4_open *open)
4220 {
4221         __be32 status;
4222         unsigned char old_deny_bmap = stp->st_deny_bmap;
4223
4224         if (!test_access(open->op_share_access, stp))
4225                 return nfs4_get_vfs_file(rqstp, fp, cur_fh, stp, open);
4226
4227         /* test and set deny mode */
4228         spin_lock(&fp->fi_lock);
4229         status = nfs4_file_check_deny(fp, open->op_share_deny);
4230         if (status == nfs_ok) {
4231                 set_deny(open->op_share_deny, stp);
4232                 fp->fi_share_deny |=
4233                                 (open->op_share_deny & NFS4_SHARE_DENY_BOTH);
4234         }
4235         spin_unlock(&fp->fi_lock);
4236
4237         if (status != nfs_ok)
4238                 return status;
4239
4240         status = nfsd4_truncate(rqstp, cur_fh, open);
4241         if (status != nfs_ok)
4242                 reset_union_bmap_deny(old_deny_bmap, stp);
4243         return status;
4244 }
4245
4246 /* Should we give out recallable state?: */
4247 static bool nfsd4_cb_channel_good(struct nfs4_client *clp)
4248 {
4249         if (clp->cl_cb_state == NFSD4_CB_UP)
4250                 return true;
4251         /*
4252          * In the sessions case, since we don't have to establish a
4253          * separate connection for callbacks, we assume it's OK
4254          * until we hear otherwise:
4255          */
4256         return clp->cl_minorversion && clp->cl_cb_state == NFSD4_CB_UNKNOWN;
4257 }
4258
4259 static struct file_lock *nfs4_alloc_init_lease(struct nfs4_file *fp, int flag)
4260 {
4261         struct file_lock *fl;
4262
4263         fl = locks_alloc_lock();
4264         if (!fl)
4265                 return NULL;
4266         fl->fl_lmops = &nfsd_lease_mng_ops;
4267         fl->fl_flags = FL_DELEG;
4268         fl->fl_type = flag == NFS4_OPEN_DELEGATE_READ? F_RDLCK: F_WRLCK;
4269         fl->fl_end = OFFSET_MAX;
4270         fl->fl_owner = (fl_owner_t)fp;
4271         fl->fl_pid = current->tgid;
4272         return fl;
4273 }
4274
4275 /**
4276  * nfs4_setlease - Obtain a delegation by requesting lease from vfs layer
4277  * @dp:   a pointer to the nfs4_delegation we're adding.
4278  *
4279  * Return:
4280  *      On success: Return code will be 0 on success.
4281  *
4282  *      On error: -EAGAIN if there was an existing delegation.
4283  *                 nonzero if there is an error in other cases.
4284  *
4285  */
4286
4287 static int nfs4_setlease(struct nfs4_delegation *dp)
4288 {
4289         struct nfs4_file *fp = dp->dl_stid.sc_file;
4290         struct file_lock *fl;
4291         struct file *filp;
4292         int status = 0;
4293
4294         fl = nfs4_alloc_init_lease(fp, NFS4_OPEN_DELEGATE_READ);
4295         if (!fl)
4296                 return -ENOMEM;
4297         filp = find_readable_file(fp);
4298         if (!filp) {
4299                 /* We should always have a readable file here */
4300                 WARN_ON_ONCE(1);
4301                 locks_free_lock(fl);
4302                 return -EBADF;
4303         }
4304         fl->fl_file = filp;
4305         status = vfs_setlease(filp, fl->fl_type, &fl, NULL);
4306         if (fl)
4307                 locks_free_lock(fl);
4308         if (status)
4309                 goto out_fput;
4310         spin_lock(&state_lock);
4311         spin_lock(&fp->fi_lock);
4312         /* Did the lease get broken before we took the lock? */
4313         status = -EAGAIN;
4314         if (fp->fi_had_conflict)
4315                 goto out_unlock;
4316         /* Race breaker */
4317         if (fp->fi_deleg_file) {
4318                 status = hash_delegation_locked(dp, fp);
4319                 goto out_unlock;
4320         }
4321         fp->fi_deleg_file = filp;
4322         fp->fi_delegees = 0;
4323         status = hash_delegation_locked(dp, fp);
4324         spin_unlock(&fp->fi_lock);
4325         spin_unlock(&state_lock);
4326         if (status) {
4327                 /* Should never happen, this is a new fi_deleg_file  */
4328                 WARN_ON_ONCE(1);
4329                 goto out_fput;
4330         }
4331         return 0;
4332 out_unlock:
4333         spin_unlock(&fp->fi_lock);
4334         spin_unlock(&state_lock);
4335 out_fput:
4336         fput(filp);
4337         return status;
4338 }
4339
4340 static struct nfs4_delegation *
4341 nfs4_set_delegation(struct nfs4_client *clp, struct svc_fh *fh,
4342                     struct nfs4_file *fp, struct nfs4_clnt_odstate *odstate)
4343 {
4344         int status;
4345         struct nfs4_delegation *dp;
4346
4347         if (fp->fi_had_conflict)
4348                 return ERR_PTR(-EAGAIN);
4349
4350         spin_lock(&state_lock);
4351         spin_lock(&fp->fi_lock);
4352         status = nfs4_get_existing_delegation(clp, fp);
4353         spin_unlock(&fp->fi_lock);
4354         spin_unlock(&state_lock);
4355
4356         if (status)
4357                 return ERR_PTR(status);
4358
4359         dp = alloc_init_deleg(clp, fh, odstate);
4360         if (!dp)
4361                 return ERR_PTR(-ENOMEM);
4362
4363         get_nfs4_file(fp);
4364         spin_lock(&state_lock);
4365         spin_lock(&fp->fi_lock);
4366         dp->dl_stid.sc_file = fp;
4367         if (!fp->fi_deleg_file) {
4368                 spin_unlock(&fp->fi_lock);
4369                 spin_unlock(&state_lock);
4370                 status = nfs4_setlease(dp);
4371                 goto out;
4372         }
4373         if (fp->fi_had_conflict) {
4374                 status = -EAGAIN;
4375                 goto out_unlock;
4376         }
4377         status = hash_delegation_locked(dp, fp);
4378 out_unlock:
4379         spin_unlock(&fp->fi_lock);
4380         spin_unlock(&state_lock);
4381 out:
4382         if (status) {
4383                 put_clnt_odstate(dp->dl_clnt_odstate);
4384                 nfs4_put_stid(&dp->dl_stid);
4385                 return ERR_PTR(status);
4386         }
4387         return dp;
4388 }
4389
4390 static void nfsd4_open_deleg_none_ext(struct nfsd4_open *open, int status)
4391 {
4392         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4393         if (status == -EAGAIN)
4394                 open->op_why_no_deleg = WND4_CONTENTION;
4395         else {
4396                 open->op_why_no_deleg = WND4_RESOURCE;
4397                 switch (open->op_deleg_want) {
4398                 case NFS4_SHARE_WANT_READ_DELEG:
4399                 case NFS4_SHARE_WANT_WRITE_DELEG:
4400                 case NFS4_SHARE_WANT_ANY_DELEG:
4401                         break;
4402                 case NFS4_SHARE_WANT_CANCEL:
4403                         open->op_why_no_deleg = WND4_CANCELLED;
4404                         break;
4405                 case NFS4_SHARE_WANT_NO_DELEG:
4406                         WARN_ON_ONCE(1);
4407                 }
4408         }
4409 }
4410
4411 /*
4412  * Attempt to hand out a delegation.
4413  *
4414  * Note we don't support write delegations, and won't until the vfs has
4415  * proper support for them.
4416  */
4417 static void
4418 nfs4_open_delegation(struct svc_fh *fh, struct nfsd4_open *open,
4419                         struct nfs4_ol_stateid *stp)
4420 {
4421         struct nfs4_delegation *dp;
4422         struct nfs4_openowner *oo = openowner(stp->st_stateowner);
4423         struct nfs4_client *clp = stp->st_stid.sc_client;
4424         int cb_up;
4425         int status = 0;
4426
4427         cb_up = nfsd4_cb_channel_good(oo->oo_owner.so_client);
4428         open->op_recall = 0;
4429         switch (open->op_claim_type) {
4430                 case NFS4_OPEN_CLAIM_PREVIOUS:
4431                         if (!cb_up)
4432                                 open->op_recall = 1;
4433                         if (open->op_delegate_type != NFS4_OPEN_DELEGATE_READ)
4434                                 goto out_no_deleg;
4435                         break;
4436                 case NFS4_OPEN_CLAIM_NULL:
4437                 case NFS4_OPEN_CLAIM_FH:
4438                         /*
4439                          * Let's not give out any delegations till everyone's
4440                          * had the chance to reclaim theirs, *and* until
4441                          * NLM locks have all been reclaimed:
4442                          */
4443                         if (locks_in_grace(clp->net))
4444                                 goto out_no_deleg;
4445                         if (!cb_up || !(oo->oo_flags & NFS4_OO_CONFIRMED))
4446                                 goto out_no_deleg;
4447                         /*
4448                          * Also, if the file was opened for write or
4449                          * create, there's a good chance the client's
4450                          * about to write to it, resulting in an
4451                          * immediate recall (since we don't support
4452                          * write delegations):
4453                          */
4454                         if (open->op_share_access & NFS4_SHARE_ACCESS_WRITE)
4455                                 goto out_no_deleg;
4456                         if (open->op_create == NFS4_OPEN_CREATE)
4457                                 goto out_no_deleg;
4458                         break;
4459                 default:
4460                         goto out_no_deleg;
4461         }
4462         dp = nfs4_set_delegation(clp, fh, stp->st_stid.sc_file, stp->st_clnt_odstate);
4463         if (IS_ERR(dp))
4464                 goto out_no_deleg;
4465
4466         memcpy(&open->op_delegate_stateid, &dp->dl_stid.sc_stateid, sizeof(dp->dl_stid.sc_stateid));
4467
4468         dprintk("NFSD: delegation stateid=" STATEID_FMT "\n",
4469                 STATEID_VAL(&dp->dl_stid.sc_stateid));
4470         open->op_delegate_type = NFS4_OPEN_DELEGATE_READ;
4471         nfs4_put_stid(&dp->dl_stid);
4472         return;
4473 out_no_deleg:
4474         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE;
4475         if (open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS &&
4476             open->op_delegate_type != NFS4_OPEN_DELEGATE_NONE) {
4477                 dprintk("NFSD: WARNING: refusing delegation reclaim\n");
4478                 open->op_recall = 1;
4479         }
4480
4481         /* 4.1 client asking for a delegation? */
4482         if (open->op_deleg_want)
4483                 nfsd4_open_deleg_none_ext(open, status);
4484         return;
4485 }
4486
4487 static void nfsd4_deleg_xgrade_none_ext(struct nfsd4_open *open,
4488                                         struct nfs4_delegation *dp)
4489 {
4490         if (open->op_deleg_want == NFS4_SHARE_WANT_READ_DELEG &&
4491             dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4492                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4493                 open->op_why_no_deleg = WND4_NOT_SUPP_DOWNGRADE;
4494         } else if (open->op_deleg_want == NFS4_SHARE_WANT_WRITE_DELEG &&
4495                    dp->dl_type == NFS4_OPEN_DELEGATE_WRITE) {
4496                 open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4497                 open->op_why_no_deleg = WND4_NOT_SUPP_UPGRADE;
4498         }
4499         /* Otherwise the client must be confused wanting a delegation
4500          * it already has, therefore we don't return
4501          * NFS4_OPEN_DELEGATE_NONE_EXT and reason.
4502          */
4503 }
4504
4505 __be32
4506 nfsd4_process_open2(struct svc_rqst *rqstp, struct svc_fh *current_fh, struct nfsd4_open *open)
4507 {
4508         struct nfsd4_compoundres *resp = rqstp->rq_resp;
4509         struct nfs4_client *cl = open->op_openowner->oo_owner.so_client;
4510         struct nfs4_file *fp = NULL;
4511         struct nfs4_ol_stateid *stp = NULL;
4512         struct nfs4_delegation *dp = NULL;
4513         __be32 status;
4514         bool new_stp = false;
4515
4516         /*
4517          * Lookup file; if found, lookup stateid and check open request,
4518          * and check for delegations in the process of being recalled.
4519          * If not found, create the nfs4_file struct
4520          */
4521         fp = find_or_add_file(open->op_file, &current_fh->fh_handle);
4522         if (fp != open->op_file) {
4523                 status = nfs4_check_deleg(cl, open, &dp);
4524                 if (status)
4525                         goto out;
4526                 stp = nfsd4_find_and_lock_existing_open(fp, open);
4527         } else {
4528                 open->op_file = NULL;
4529                 status = nfserr_bad_stateid;
4530                 if (nfsd4_is_deleg_cur(open))
4531                         goto out;
4532         }
4533
4534         if (!stp) {
4535                 stp = init_open_stateid(fp, open);
4536                 if (!open->op_stp)
4537                         new_stp = true;
4538         }
4539
4540         /*
4541          * OPEN the file, or upgrade an existing OPEN.
4542          * If truncate fails, the OPEN fails.
4543          *
4544          * stp is already locked.
4545          */
4546         if (!new_stp) {
4547                 /* Stateid was found, this is an OPEN upgrade */
4548                 status = nfs4_upgrade_open(rqstp, fp, current_fh, stp, open);
4549                 if (status) {
4550                         mutex_unlock(&stp->st_mutex);
4551                         goto out;
4552                 }
4553         } else {
4554                 status = nfs4_get_vfs_file(rqstp, fp, current_fh, stp, open);
4555                 if (status) {
4556                         stp->st_stid.sc_type = NFS4_CLOSED_STID;
4557                         release_open_stateid(stp);
4558                         mutex_unlock(&stp->st_mutex);
4559                         goto out;
4560                 }
4561
4562                 stp->st_clnt_odstate = find_or_hash_clnt_odstate(fp,
4563                                                         open->op_odstate);
4564                 if (stp->st_clnt_odstate == open->op_odstate)
4565                         open->op_odstate = NULL;
4566         }
4567
4568         nfs4_inc_and_copy_stateid(&open->op_stateid, &stp->st_stid);
4569         mutex_unlock(&stp->st_mutex);
4570
4571         if (nfsd4_has_session(&resp->cstate)) {
4572                 if (open->op_deleg_want & NFS4_SHARE_WANT_NO_DELEG) {
4573                         open->op_delegate_type = NFS4_OPEN_DELEGATE_NONE_EXT;
4574                         open->op_why_no_deleg = WND4_NOT_WANTED;
4575                         goto nodeleg;
4576                 }
4577         }
4578
4579         /*
4580         * Attempt to hand out a delegation. No error return, because the
4581         * OPEN succeeds even if we fail.
4582         */
4583         nfs4_open_delegation(current_fh, open, stp);
4584 nodeleg:
4585         status = nfs_ok;
4586
4587         dprintk("%s: stateid=" STATEID_FMT "\n", __func__,
4588                 STATEID_VAL(&stp->st_stid.sc_stateid));
4589 out:
4590         /* 4.1 client trying to upgrade/downgrade delegation? */
4591         if (open->op_delegate_type == NFS4_OPEN_DELEGATE_NONE && dp &&
4592             open->op_deleg_want)
4593                 nfsd4_deleg_xgrade_none_ext(open, dp);
4594
4595         if (fp)
4596                 put_nfs4_file(fp);
4597         if (status == 0 && open->op_claim_type == NFS4_OPEN_CLAIM_PREVIOUS)
4598                 open->op_openowner->oo_flags |= NFS4_OO_CONFIRMED;
4599         /*
4600         * To finish the open response, we just need to set the rflags.
4601         */
4602         open->op_rflags = NFS4_OPEN_RESULT_LOCKTYPE_POSIX;
4603         if (nfsd4_has_session(&resp->cstate))
4604                 open->op_rflags |= NFS4_OPEN_RESULT_MAY_NOTIFY_LOCK;
4605         else if (!(open->op_openowner->oo_flags & NFS4_OO_CONFIRMED))
4606                 open->op_rflags |= NFS4_OPEN_RESULT_CONFIRM;
4607
4608         if (dp)
4609                 nfs4_put_stid(&dp->dl_stid);
4610         if (stp)
4611                 nfs4_put_stid(&stp->st_stid);
4612
4613         return status;
4614 }
4615
4616 void nfsd4_cleanup_open_state(struct nfsd4_compound_state *cstate,
4617                               struct nfsd4_open *open)
4618 {
4619         if (open->op_openowner) {
4620                 struct nfs4_stateowner *so = &open->op_openowner->oo_owner;
4621
4622                 nfsd4_cstate_assign_replay(cstate, so);
4623                 nfs4_put_stateowner(so);
4624         }
4625         if (open->op_file)
4626                 kmem_cache_free(file_slab, open->op_file);
4627         if (open->op_stp)
4628                 nfs4_put_stid(&open->op_stp->st_stid);
4629         if (open->op_odstate)
4630                 kmem_cache_free(odstate_slab, open->op_odstate);
4631 }
4632
4633 __be32
4634 nfsd4_renew(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
4635             union nfsd4_op_u *u)
4636 {
4637         clientid_t *clid = &u->renew;
4638         struct nfs4_client *clp;
4639         __be32 status;
4640         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
4641
4642         dprintk("process_renew(%08x/%08x): starting\n",
4643                         clid->cl_boot, clid->cl_id);
4644         status = lookup_clientid(clid, cstate, nn);
4645         if (status)
4646                 goto out;
4647         clp = cstate->clp;
4648         status = nfserr_cb_path_down;
4649         if (!list_empty(&clp->cl_delegations)
4650                         && clp->cl_cb_state != NFSD4_CB_UP)
4651                 goto out;
4652         status = nfs_ok;
4653 out:
4654         return status;
4655 }
4656
4657 void
4658 nfsd4_end_grace(struct nfsd_net *nn)
4659 {
4660         /* do nothing if grace period already ended */
4661         if (nn->grace_ended)
4662                 return;
4663
4664         dprintk("NFSD: end of grace period\n");
4665         nn->grace_ended = true;
4666         /*
4667          * If the server goes down again right now, an NFSv4
4668          * client will still be allowed to reclaim after it comes back up,
4669          * even if it hasn't yet had a chance to reclaim state this time.
4670          *
4671          */
4672         nfsd4_record_grace_done(nn);
4673         /*
4674          * At this point, NFSv4 clients can still reclaim.  But if the
4675          * server crashes, any that have not yet reclaimed will be out
4676          * of luck on the next boot.
4677          *
4678          * (NFSv4.1+ clients are considered to have reclaimed once they
4679          * call RECLAIM_COMPLETE.  NFSv4.0 clients are considered to
4680          * have reclaimed after their first OPEN.)
4681          */
4682         locks_end_grace(&nn->nfsd4_manager);
4683         /*
4684          * At this point, and once lockd and/or any other containers
4685          * exit their grace period, further reclaims will fail and
4686          * regular locking can resume.
4687          */
4688 }
4689
4690 static time_t
4691 nfs4_laundromat(struct nfsd_net *nn)
4692 {
4693         struct nfs4_client *clp;
4694         struct nfs4_openowner *oo;
4695         struct nfs4_delegation *dp;
4696         struct nfs4_ol_stateid *stp;
4697         struct nfsd4_blocked_lock *nbl;
4698         struct list_head *pos, *next, reaplist;
4699         time_t cutoff = get_seconds() - nn->nfsd4_lease;
4700         time_t t, new_timeo = nn->nfsd4_lease;
4701
4702         dprintk("NFSD: laundromat service - starting\n");
4703         nfsd4_end_grace(nn);
4704         INIT_LIST_HEAD(&reaplist);
4705         spin_lock(&nn->client_lock);
4706         list_for_each_safe(pos, next, &nn->client_lru) {
4707                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4708                 if (time_after((unsigned long)clp->cl_time, (unsigned long)cutoff)) {
4709                         t = clp->cl_time - cutoff;
4710                         new_timeo = min(new_timeo, t);
4711                         break;
4712                 }
4713                 if (mark_client_expired_locked(clp)) {
4714                         dprintk("NFSD: client in use (clientid %08x)\n",
4715                                 clp->cl_clientid.cl_id);
4716                         continue;
4717                 }
4718                 list_add(&clp->cl_lru, &reaplist);
4719         }
4720         spin_unlock(&nn->client_lock);
4721         list_for_each_safe(pos, next, &reaplist) {
4722                 clp = list_entry(pos, struct nfs4_client, cl_lru);
4723                 dprintk("NFSD: purging unused client (clientid %08x)\n",
4724                         clp->cl_clientid.cl_id);
4725                 list_del_init(&clp->cl_lru);
4726                 expire_client(clp);
4727         }
4728         spin_lock(&state_lock);
4729         list_for_each_safe(pos, next, &nn->del_recall_lru) {
4730                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
4731                 if (time_after((unsigned long)dp->dl_time, (unsigned long)cutoff)) {
4732                         t = dp->dl_time - cutoff;
4733                         new_timeo = min(new_timeo, t);
4734                         break;
4735                 }
4736                 WARN_ON(!unhash_delegation_locked(dp));
4737                 list_add(&dp->dl_recall_lru, &reaplist);
4738         }
4739         spin_unlock(&state_lock);
4740         while (!list_empty(&reaplist)) {
4741                 dp = list_first_entry(&reaplist, struct nfs4_delegation,
4742                                         dl_recall_lru);
4743                 list_del_init(&dp->dl_recall_lru);
4744                 revoke_delegation(dp);
4745         }
4746
4747         spin_lock(&nn->client_lock);
4748         while (!list_empty(&nn->close_lru)) {
4749                 oo = list_first_entry(&nn->close_lru, struct nfs4_openowner,
4750                                         oo_close_lru);
4751                 if (time_after((unsigned long)oo->oo_time,
4752                                (unsigned long)cutoff)) {
4753                         t = oo->oo_time - cutoff;
4754                         new_timeo = min(new_timeo, t);
4755                         break;
4756                 }
4757                 list_del_init(&oo->oo_close_lru);
4758                 stp = oo->oo_last_closed_stid;
4759                 oo->oo_last_closed_stid = NULL;
4760                 spin_unlock(&nn->client_lock);
4761                 nfs4_put_stid(&stp->st_stid);
4762                 spin_lock(&nn->client_lock);
4763         }
4764         spin_unlock(&nn->client_lock);
4765
4766         /*
4767          * It's possible for a client to try and acquire an already held lock
4768          * that is being held for a long time, and then lose interest in it.
4769          * So, we clean out any un-revisited request after a lease period
4770          * under the assumption that the client is no longer interested.
4771          *
4772          * RFC5661, sec. 9.6 states that the client must not rely on getting
4773          * notifications and must continue to poll for locks, even when the
4774          * server supports them. Thus this shouldn't lead to clients blocking
4775          * indefinitely once the lock does become free.
4776          */
4777         BUG_ON(!list_empty(&reaplist));
4778         spin_lock(&nn->blocked_locks_lock);
4779         while (!list_empty(&nn->blocked_locks_lru)) {
4780                 nbl = list_first_entry(&nn->blocked_locks_lru,
4781                                         struct nfsd4_blocked_lock, nbl_lru);
4782                 if (time_after((unsigned long)nbl->nbl_time,
4783                                (unsigned long)cutoff)) {
4784                         t = nbl->nbl_time - cutoff;
4785                         new_timeo = min(new_timeo, t);
4786                         break;
4787                 }
4788                 list_move(&nbl->nbl_lru, &reaplist);
4789                 list_del_init(&nbl->nbl_list);
4790         }
4791         spin_unlock(&nn->blocked_locks_lock);
4792
4793         while (!list_empty(&reaplist)) {
4794                 nbl = list_first_entry(&reaplist,
4795                                         struct nfsd4_blocked_lock, nbl_lru);
4796                 list_del_init(&nbl->nbl_lru);
4797                 posix_unblock_lock(&nbl->nbl_lock);
4798                 free_blocked_lock(nbl);
4799         }
4800
4801         new_timeo = max_t(time_t, new_timeo, NFSD_LAUNDROMAT_MINTIMEOUT);
4802         return new_timeo;
4803 }
4804
4805 static struct workqueue_struct *laundry_wq;
4806 static void laundromat_main(struct work_struct *);
4807
4808 static void
4809 laundromat_main(struct work_struct *laundry)
4810 {
4811         time_t t;
4812         struct delayed_work *dwork = to_delayed_work(laundry);
4813         struct nfsd_net *nn = container_of(dwork, struct nfsd_net,
4814                                            laundromat_work);
4815
4816         t = nfs4_laundromat(nn);
4817         dprintk("NFSD: laundromat_main - sleeping for %ld seconds\n", t);
4818         queue_delayed_work(laundry_wq, &nn->laundromat_work, t*HZ);
4819 }
4820
4821 static inline __be32 nfs4_check_fh(struct svc_fh *fhp, struct nfs4_stid *stp)
4822 {
4823         if (!fh_match(&fhp->fh_handle, &stp->sc_file->fi_fhandle))
4824                 return nfserr_bad_stateid;
4825         return nfs_ok;
4826 }
4827
4828 static inline int
4829 access_permit_read(struct nfs4_ol_stateid *stp)
4830 {
4831         return test_access(NFS4_SHARE_ACCESS_READ, stp) ||
4832                 test_access(NFS4_SHARE_ACCESS_BOTH, stp) ||
4833                 test_access(NFS4_SHARE_ACCESS_WRITE, stp);
4834 }
4835
4836 static inline int
4837 access_permit_write(struct nfs4_ol_stateid *stp)
4838 {
4839         return test_access(NFS4_SHARE_ACCESS_WRITE, stp) ||
4840                 test_access(NFS4_SHARE_ACCESS_BOTH, stp);
4841 }
4842
4843 static
4844 __be32 nfs4_check_openmode(struct nfs4_ol_stateid *stp, int flags)
4845 {
4846         __be32 status = nfserr_openmode;
4847
4848         /* For lock stateid's, we test the parent open, not the lock: */
4849         if (stp->st_openstp)
4850                 stp = stp->st_openstp;
4851         if ((flags & WR_STATE) && !access_permit_write(stp))
4852                 goto out;
4853         if ((flags & RD_STATE) && !access_permit_read(stp))
4854                 goto out;
4855         status = nfs_ok;
4856 out:
4857         return status;
4858 }
4859
4860 static inline __be32
4861 check_special_stateids(struct net *net, svc_fh *current_fh, stateid_t *stateid, int flags)
4862 {
4863         if (ONE_STATEID(stateid) && (flags & RD_STATE))
4864                 return nfs_ok;
4865         else if (opens_in_grace(net)) {
4866                 /* Answer in remaining cases depends on existence of
4867                  * conflicting state; so we must wait out the grace period. */
4868                 return nfserr_grace;
4869         } else if (flags & WR_STATE)
4870                 return nfs4_share_conflict(current_fh,
4871                                 NFS4_SHARE_DENY_WRITE);
4872         else /* (flags & RD_STATE) && ZERO_STATEID(stateid) */
4873                 return nfs4_share_conflict(current_fh,
4874                                 NFS4_SHARE_DENY_READ);
4875 }
4876
4877 /*
4878  * Allow READ/WRITE during grace period on recovered state only for files
4879  * that are not able to provide mandatory locking.
4880  */
4881 static inline int
4882 grace_disallows_io(struct net *net, struct inode *inode)
4883 {
4884         return opens_in_grace(net) && mandatory_lock(inode);
4885 }
4886
4887 static __be32 check_stateid_generation(stateid_t *in, stateid_t *ref, bool has_session)
4888 {
4889         /*
4890          * When sessions are used the stateid generation number is ignored
4891          * when it is zero.
4892          */
4893         if (has_session && in->si_generation == 0)
4894                 return nfs_ok;
4895
4896         if (in->si_generation == ref->si_generation)
4897                 return nfs_ok;
4898
4899         /* If the client sends us a stateid from the future, it's buggy: */
4900         if (nfsd4_stateid_generation_after(in, ref))
4901                 return nfserr_bad_stateid;
4902         /*
4903          * However, we could see a stateid from the past, even from a
4904          * non-buggy client.  For example, if the client sends a lock
4905          * while some IO is outstanding, the lock may bump si_generation
4906          * while the IO is still in flight.  The client could avoid that
4907          * situation by waiting for responses on all the IO requests,
4908          * but better performance may result in retrying IO that
4909          * receives an old_stateid error if requests are rarely
4910          * reordered in flight:
4911          */
4912         return nfserr_old_stateid;
4913 }
4914
4915 static __be32 nfsd4_stid_check_stateid_generation(stateid_t *in, struct nfs4_stid *s, bool has_session)
4916 {
4917         __be32 ret;
4918
4919         spin_lock(&s->sc_lock);
4920         ret = nfsd4_verify_open_stid(s);
4921         if (ret == nfs_ok)
4922                 ret = check_stateid_generation(in, &s->sc_stateid, has_session);
4923         spin_unlock(&s->sc_lock);
4924         return ret;
4925 }
4926
4927 static __be32 nfsd4_check_openowner_confirmed(struct nfs4_ol_stateid *ols)
4928 {
4929         if (ols->st_stateowner->so_is_open_owner &&
4930             !(openowner(ols->st_stateowner)->oo_flags & NFS4_OO_CONFIRMED))
4931                 return nfserr_bad_stateid;
4932         return nfs_ok;
4933 }
4934
4935 static __be32 nfsd4_validate_stateid(struct nfs4_client *cl, stateid_t *stateid)
4936 {
4937         struct nfs4_stid *s;
4938         __be32 status = nfserr_bad_stateid;
4939
4940         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
4941                 CLOSE_STATEID(stateid))
4942                 return status;
4943         /* Client debugging aid. */
4944         if (!same_clid(&stateid->si_opaque.so_clid, &cl->cl_clientid)) {
4945                 char addr_str[INET6_ADDRSTRLEN];
4946                 rpc_ntop((struct sockaddr *)&cl->cl_addr, addr_str,
4947                                  sizeof(addr_str));
4948                 pr_warn_ratelimited("NFSD: client %s testing state ID "
4949                                         "with incorrect client ID\n", addr_str);
4950                 return status;
4951         }
4952         spin_lock(&cl->cl_lock);
4953         s = find_stateid_locked(cl, stateid);
4954         if (!s)
4955                 goto out_unlock;
4956         status = nfsd4_stid_check_stateid_generation(stateid, s, 1);
4957         if (status)
4958                 goto out_unlock;
4959         switch (s->sc_type) {
4960         case NFS4_DELEG_STID:
4961                 status = nfs_ok;
4962                 break;
4963         case NFS4_REVOKED_DELEG_STID:
4964                 status = nfserr_deleg_revoked;
4965                 break;
4966         case NFS4_OPEN_STID:
4967         case NFS4_LOCK_STID:
4968                 status = nfsd4_check_openowner_confirmed(openlockstateid(s));
4969                 break;
4970         default:
4971                 printk("unknown stateid type %x\n", s->sc_type);
4972                 /* Fallthrough */
4973         case NFS4_CLOSED_STID:
4974         case NFS4_CLOSED_DELEG_STID:
4975                 status = nfserr_bad_stateid;
4976         }
4977 out_unlock:
4978         spin_unlock(&cl->cl_lock);
4979         return status;
4980 }
4981
4982 __be32
4983 nfsd4_lookup_stateid(struct nfsd4_compound_state *cstate,
4984                      stateid_t *stateid, unsigned char typemask,
4985                      struct nfs4_stid **s, struct nfsd_net *nn)
4986 {
4987         __be32 status;
4988         bool return_revoked = false;
4989
4990         /*
4991          *  only return revoked delegations if explicitly asked.
4992          *  otherwise we report revoked or bad_stateid status.
4993          */
4994         if (typemask & NFS4_REVOKED_DELEG_STID)
4995                 return_revoked = true;
4996         else if (typemask & NFS4_DELEG_STID)
4997                 typemask |= NFS4_REVOKED_DELEG_STID;
4998
4999         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid) ||
5000                 CLOSE_STATEID(stateid))
5001                 return nfserr_bad_stateid;
5002         status = lookup_clientid(&stateid->si_opaque.so_clid, cstate, nn);
5003         if (status == nfserr_stale_clientid) {
5004                 if (cstate->session)
5005                         return nfserr_bad_stateid;
5006                 return nfserr_stale_stateid;
5007         }
5008         if (status)
5009                 return status;
5010         *s = find_stateid_by_type(cstate->clp, stateid, typemask);
5011         if (!*s)
5012                 return nfserr_bad_stateid;
5013         if (((*s)->sc_type == NFS4_REVOKED_DELEG_STID) && !return_revoked) {
5014                 nfs4_put_stid(*s);
5015                 if (cstate->minorversion)
5016                         return nfserr_deleg_revoked;
5017                 return nfserr_bad_stateid;
5018         }
5019         return nfs_ok;
5020 }
5021
5022 static struct file *
5023 nfs4_find_file(struct nfs4_stid *s, int flags)
5024 {
5025         if (!s)
5026                 return NULL;
5027
5028         switch (s->sc_type) {
5029         case NFS4_DELEG_STID:
5030                 if (WARN_ON_ONCE(!s->sc_file->fi_deleg_file))
5031                         return NULL;
5032                 return get_file(s->sc_file->fi_deleg_file);
5033         case NFS4_OPEN_STID:
5034         case NFS4_LOCK_STID:
5035                 if (flags & RD_STATE)
5036                         return find_readable_file(s->sc_file);
5037                 else
5038                         return find_writeable_file(s->sc_file);
5039                 break;
5040         }
5041
5042         return NULL;
5043 }
5044
5045 static __be32
5046 nfs4_check_olstateid(struct svc_fh *fhp, struct nfs4_ol_stateid *ols, int flags)
5047 {
5048         __be32 status;
5049
5050         status = nfsd4_check_openowner_confirmed(ols);
5051         if (status)
5052                 return status;
5053         return nfs4_check_openmode(ols, flags);
5054 }
5055
5056 static __be32
5057 nfs4_check_file(struct svc_rqst *rqstp, struct svc_fh *fhp, struct nfs4_stid *s,
5058                 struct file **filpp, bool *tmp_file, int flags)
5059 {
5060         int acc = (flags & RD_STATE) ? NFSD_MAY_READ : NFSD_MAY_WRITE;
5061         struct file *file;
5062         __be32 status;
5063
5064         file = nfs4_find_file(s, flags);
5065         if (file) {
5066                 status = nfsd_permission(rqstp, fhp->fh_export, fhp->fh_dentry,
5067                                 acc | NFSD_MAY_OWNER_OVERRIDE);
5068                 if (status) {
5069                         fput(file);
5070                         return status;
5071                 }
5072
5073                 *filpp = file;
5074         } else {
5075                 status = nfsd_open(rqstp, fhp, S_IFREG, acc, filpp);
5076                 if (status)
5077                         return status;
5078
5079                 if (tmp_file)
5080                         *tmp_file = true;
5081         }
5082
5083         return 0;
5084 }
5085
5086 /*
5087  * Checks for stateid operations
5088  */
5089 __be32
5090 nfs4_preprocess_stateid_op(struct svc_rqst *rqstp,
5091                 struct nfsd4_compound_state *cstate, struct svc_fh *fhp,
5092                 stateid_t *stateid, int flags, struct file **filpp, bool *tmp_file)
5093 {
5094         struct inode *ino = d_inode(fhp->fh_dentry);
5095         struct net *net = SVC_NET(rqstp);
5096         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5097         struct nfs4_stid *s = NULL;
5098         __be32 status;
5099
5100         if (filpp)
5101                 *filpp = NULL;
5102         if (tmp_file)
5103                 *tmp_file = false;
5104
5105         if (grace_disallows_io(net, ino))
5106                 return nfserr_grace;
5107
5108         if (ZERO_STATEID(stateid) || ONE_STATEID(stateid)) {
5109                 status = check_special_stateids(net, fhp, stateid, flags);
5110                 goto done;
5111         }
5112
5113         status = nfsd4_lookup_stateid(cstate, stateid,
5114                                 NFS4_DELEG_STID|NFS4_OPEN_STID|NFS4_LOCK_STID,
5115                                 &s, nn);
5116         if (status)
5117                 return status;
5118         status = nfsd4_stid_check_stateid_generation(stateid, s,
5119                         nfsd4_has_session(cstate));
5120         if (status)
5121                 goto out;
5122
5123         switch (s->sc_type) {
5124         case NFS4_DELEG_STID:
5125                 status = nfs4_check_delegmode(delegstateid(s), flags);
5126                 break;
5127         case NFS4_OPEN_STID:
5128         case NFS4_LOCK_STID:
5129                 status = nfs4_check_olstateid(fhp, openlockstateid(s), flags);
5130                 break;
5131         default:
5132                 status = nfserr_bad_stateid;
5133                 break;
5134         }
5135         if (status)
5136                 goto out;
5137         status = nfs4_check_fh(fhp, s);
5138
5139 done:
5140         if (!status && filpp)
5141                 status = nfs4_check_file(rqstp, fhp, s, filpp, tmp_file, flags);
5142 out:
5143         if (s)
5144                 nfs4_put_stid(s);
5145         return status;
5146 }
5147
5148 /*
5149  * Test if the stateid is valid
5150  */
5151 __be32
5152 nfsd4_test_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5153                    union nfsd4_op_u *u)
5154 {
5155         struct nfsd4_test_stateid *test_stateid = &u->test_stateid;
5156         struct nfsd4_test_stateid_id *stateid;
5157         struct nfs4_client *cl = cstate->session->se_client;
5158
5159         list_for_each_entry(stateid, &test_stateid->ts_stateid_list, ts_id_list)
5160                 stateid->ts_id_status =
5161                         nfsd4_validate_stateid(cl, &stateid->ts_id_stateid);
5162
5163         return nfs_ok;
5164 }
5165
5166 static __be32
5167 nfsd4_free_lock_stateid(stateid_t *stateid, struct nfs4_stid *s)
5168 {
5169         struct nfs4_ol_stateid *stp = openlockstateid(s);
5170         __be32 ret;
5171
5172         ret = nfsd4_lock_ol_stateid(stp);
5173         if (ret)
5174                 goto out_put_stid;
5175
5176         ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5177         if (ret)
5178                 goto out;
5179
5180         ret = nfserr_locks_held;
5181         if (check_for_locks(stp->st_stid.sc_file,
5182                             lockowner(stp->st_stateowner)))
5183                 goto out;
5184
5185         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5186         release_lock_stateid(stp);
5187         ret = nfs_ok;
5188
5189 out:
5190         mutex_unlock(&stp->st_mutex);
5191 out_put_stid:
5192         nfs4_put_stid(s);
5193         return ret;
5194 }
5195
5196 __be32
5197 nfsd4_free_stateid(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5198                    union nfsd4_op_u *u)
5199 {
5200         struct nfsd4_free_stateid *free_stateid = &u->free_stateid;
5201         stateid_t *stateid = &free_stateid->fr_stateid;
5202         struct nfs4_stid *s;
5203         struct nfs4_delegation *dp;
5204         struct nfs4_client *cl = cstate->session->se_client;
5205         __be32 ret = nfserr_bad_stateid;
5206
5207         spin_lock(&cl->cl_lock);
5208         s = find_stateid_locked(cl, stateid);
5209         if (!s)
5210                 goto out_unlock;
5211         spin_lock(&s->sc_lock);
5212         switch (s->sc_type) {
5213         case NFS4_DELEG_STID:
5214                 ret = nfserr_locks_held;
5215                 break;
5216         case NFS4_OPEN_STID:
5217                 ret = check_stateid_generation(stateid, &s->sc_stateid, 1);
5218                 if (ret)
5219                         break;
5220                 ret = nfserr_locks_held;
5221                 break;
5222         case NFS4_LOCK_STID:
5223                 spin_unlock(&s->sc_lock);
5224                 refcount_inc(&s->sc_count);
5225                 spin_unlock(&cl->cl_lock);
5226                 ret = nfsd4_free_lock_stateid(stateid, s);
5227                 goto out;
5228         case NFS4_REVOKED_DELEG_STID:
5229                 spin_unlock(&s->sc_lock);
5230                 dp = delegstateid(s);
5231                 list_del_init(&dp->dl_recall_lru);
5232                 spin_unlock(&cl->cl_lock);
5233                 nfs4_put_stid(s);
5234                 ret = nfs_ok;
5235                 goto out;
5236         /* Default falls through and returns nfserr_bad_stateid */
5237         }
5238         spin_unlock(&s->sc_lock);
5239 out_unlock:
5240         spin_unlock(&cl->cl_lock);
5241 out:
5242         return ret;
5243 }
5244
5245 static inline int
5246 setlkflg (int type)
5247 {
5248         return (type == NFS4_READW_LT || type == NFS4_READ_LT) ?
5249                 RD_STATE : WR_STATE;
5250 }
5251
5252 static __be32 nfs4_seqid_op_checks(struct nfsd4_compound_state *cstate, stateid_t *stateid, u32 seqid, struct nfs4_ol_stateid *stp)
5253 {
5254         struct svc_fh *current_fh = &cstate->current_fh;
5255         struct nfs4_stateowner *sop = stp->st_stateowner;
5256         __be32 status;
5257
5258         status = nfsd4_check_seqid(cstate, sop, seqid);
5259         if (status)
5260                 return status;
5261         status = nfsd4_lock_ol_stateid(stp);
5262         if (status != nfs_ok)
5263                 return status;
5264         status = check_stateid_generation(stateid, &stp->st_stid.sc_stateid, nfsd4_has_session(cstate));
5265         if (status == nfs_ok)
5266                 status = nfs4_check_fh(current_fh, &stp->st_stid);
5267         if (status != nfs_ok)
5268                 mutex_unlock(&stp->st_mutex);
5269         return status;
5270 }
5271
5272 /*
5273  * Checks for sequence id mutating operations.
5274  */
5275 static __be32
5276 nfs4_preprocess_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5277                          stateid_t *stateid, char typemask,
5278                          struct nfs4_ol_stateid **stpp,
5279                          struct nfsd_net *nn)
5280 {
5281         __be32 status;
5282         struct nfs4_stid *s;
5283         struct nfs4_ol_stateid *stp = NULL;
5284
5285         dprintk("NFSD: %s: seqid=%d stateid = " STATEID_FMT "\n", __func__,
5286                 seqid, STATEID_VAL(stateid));
5287
5288         *stpp = NULL;
5289         status = nfsd4_lookup_stateid(cstate, stateid, typemask, &s, nn);
5290         if (status)
5291                 return status;
5292         stp = openlockstateid(s);
5293         nfsd4_cstate_assign_replay(cstate, stp->st_stateowner);
5294
5295         status = nfs4_seqid_op_checks(cstate, stateid, seqid, stp);
5296         if (!status)
5297                 *stpp = stp;
5298         else
5299                 nfs4_put_stid(&stp->st_stid);
5300         return status;
5301 }
5302
5303 static __be32 nfs4_preprocess_confirmed_seqid_op(struct nfsd4_compound_state *cstate, u32 seqid,
5304                                                  stateid_t *stateid, struct nfs4_ol_stateid **stpp, struct nfsd_net *nn)
5305 {
5306         __be32 status;
5307         struct nfs4_openowner *oo;
5308         struct nfs4_ol_stateid *stp;
5309
5310         status = nfs4_preprocess_seqid_op(cstate, seqid, stateid,
5311                                                 NFS4_OPEN_STID, &stp, nn);
5312         if (status)
5313                 return status;
5314         oo = openowner(stp->st_stateowner);
5315         if (!(oo->oo_flags & NFS4_OO_CONFIRMED)) {
5316                 mutex_unlock(&stp->st_mutex);
5317                 nfs4_put_stid(&stp->st_stid);
5318                 return nfserr_bad_stateid;
5319         }
5320         *stpp = stp;
5321         return nfs_ok;
5322 }
5323
5324 __be32
5325 nfsd4_open_confirm(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5326                    union nfsd4_op_u *u)
5327 {
5328         struct nfsd4_open_confirm *oc = &u->open_confirm;
5329         __be32 status;
5330         struct nfs4_openowner *oo;
5331         struct nfs4_ol_stateid *stp;
5332         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5333
5334         dprintk("NFSD: nfsd4_open_confirm on file %pd\n",
5335                         cstate->current_fh.fh_dentry);
5336
5337         status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0);
5338         if (status)
5339                 return status;
5340
5341         status = nfs4_preprocess_seqid_op(cstate,
5342                                         oc->oc_seqid, &oc->oc_req_stateid,
5343                                         NFS4_OPEN_STID, &stp, nn);
5344         if (status)
5345                 goto out;
5346         oo = openowner(stp->st_stateowner);
5347         status = nfserr_bad_stateid;
5348         if (oo->oo_flags & NFS4_OO_CONFIRMED) {
5349                 mutex_unlock(&stp->st_mutex);
5350                 goto put_stateid;
5351         }
5352         oo->oo_flags |= NFS4_OO_CONFIRMED;
5353         nfs4_inc_and_copy_stateid(&oc->oc_resp_stateid, &stp->st_stid);
5354         mutex_unlock(&stp->st_mutex);
5355         dprintk("NFSD: %s: success, seqid=%d stateid=" STATEID_FMT "\n",
5356                 __func__, oc->oc_seqid, STATEID_VAL(&stp->st_stid.sc_stateid));
5357
5358         nfsd4_client_record_create(oo->oo_owner.so_client);
5359         status = nfs_ok;
5360 put_stateid:
5361         nfs4_put_stid(&stp->st_stid);
5362 out:
5363         nfsd4_bump_seqid(cstate, status);
5364         return status;
5365 }
5366
5367 static inline void nfs4_stateid_downgrade_bit(struct nfs4_ol_stateid *stp, u32 access)
5368 {
5369         if (!test_access(access, stp))
5370                 return;
5371         nfs4_file_put_access(stp->st_stid.sc_file, access);
5372         clear_access(access, stp);
5373 }
5374
5375 static inline void nfs4_stateid_downgrade(struct nfs4_ol_stateid *stp, u32 to_access)
5376 {
5377         switch (to_access) {
5378         case NFS4_SHARE_ACCESS_READ:
5379                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_WRITE);
5380                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5381                 break;
5382         case NFS4_SHARE_ACCESS_WRITE:
5383                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_READ);
5384                 nfs4_stateid_downgrade_bit(stp, NFS4_SHARE_ACCESS_BOTH);
5385                 break;
5386         case NFS4_SHARE_ACCESS_BOTH:
5387                 break;
5388         default:
5389                 WARN_ON_ONCE(1);
5390         }
5391 }
5392
5393 __be32
5394 nfsd4_open_downgrade(struct svc_rqst *rqstp,
5395                      struct nfsd4_compound_state *cstate, union nfsd4_op_u *u)
5396 {
5397         struct nfsd4_open_downgrade *od = &u->open_downgrade;
5398         __be32 status;
5399         struct nfs4_ol_stateid *stp;
5400         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5401
5402         dprintk("NFSD: nfsd4_open_downgrade on file %pd\n",
5403                         cstate->current_fh.fh_dentry);
5404
5405         /* We don't yet support WANT bits: */
5406         if (od->od_deleg_want)
5407                 dprintk("NFSD: %s: od_deleg_want=0x%x ignored\n", __func__,
5408                         od->od_deleg_want);
5409
5410         status = nfs4_preprocess_confirmed_seqid_op(cstate, od->od_seqid,
5411                                         &od->od_stateid, &stp, nn);
5412         if (status)
5413                 goto out;
5414         status = nfserr_inval;
5415         if (!test_access(od->od_share_access, stp)) {
5416                 dprintk("NFSD: access not a subset of current bitmap: 0x%hhx, input access=%08x\n",
5417                         stp->st_access_bmap, od->od_share_access);
5418                 goto put_stateid;
5419         }
5420         if (!test_deny(od->od_share_deny, stp)) {
5421                 dprintk("NFSD: deny not a subset of current bitmap: 0x%hhx, input deny=%08x\n",
5422                         stp->st_deny_bmap, od->od_share_deny);
5423                 goto put_stateid;
5424         }
5425         nfs4_stateid_downgrade(stp, od->od_share_access);
5426         reset_union_bmap_deny(od->od_share_deny, stp);
5427         nfs4_inc_and_copy_stateid(&od->od_stateid, &stp->st_stid);
5428         status = nfs_ok;
5429 put_stateid:
5430         mutex_unlock(&stp->st_mutex);
5431         nfs4_put_stid(&stp->st_stid);
5432 out:
5433         nfsd4_bump_seqid(cstate, status);
5434         return status;
5435 }
5436
5437 static void nfsd4_close_open_stateid(struct nfs4_ol_stateid *s)
5438 {
5439         struct nfs4_client *clp = s->st_stid.sc_client;
5440         bool unhashed;
5441         LIST_HEAD(reaplist);
5442
5443         spin_lock(&clp->cl_lock);
5444         unhashed = unhash_open_stateid(s, &reaplist);
5445
5446         if (clp->cl_minorversion) {
5447                 if (unhashed)
5448                         put_ol_stateid_locked(s, &reaplist);
5449                 spin_unlock(&clp->cl_lock);
5450                 free_ol_stateid_reaplist(&reaplist);
5451         } else {
5452                 spin_unlock(&clp->cl_lock);
5453                 free_ol_stateid_reaplist(&reaplist);
5454                 if (unhashed)
5455                         move_to_close_lru(s, clp->net);
5456         }
5457 }
5458
5459 /*
5460  * nfs4_unlock_state() called after encode
5461  */
5462 __be32
5463 nfsd4_close(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5464                 union nfsd4_op_u *u)
5465 {
5466         struct nfsd4_close *close = &u->close;
5467         __be32 status;
5468         struct nfs4_ol_stateid *stp;
5469         struct net *net = SVC_NET(rqstp);
5470         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5471
5472         dprintk("NFSD: nfsd4_close on file %pd\n",
5473                         cstate->current_fh.fh_dentry);
5474
5475         status = nfs4_preprocess_seqid_op(cstate, close->cl_seqid,
5476                                         &close->cl_stateid,
5477                                         NFS4_OPEN_STID|NFS4_CLOSED_STID,
5478                                         &stp, nn);
5479         nfsd4_bump_seqid(cstate, status);
5480         if (status)
5481                 goto out;
5482
5483         stp->st_stid.sc_type = NFS4_CLOSED_STID;
5484         nfs4_inc_and_copy_stateid(&close->cl_stateid, &stp->st_stid);
5485
5486         nfsd4_close_open_stateid(stp);
5487         mutex_unlock(&stp->st_mutex);
5488
5489         /* See RFC5661 sectionm 18.2.4 */
5490         if (stp->st_stid.sc_client->cl_minorversion)
5491                 memcpy(&close->cl_stateid, &close_stateid,
5492                                 sizeof(close->cl_stateid));
5493
5494         /* put reference from nfs4_preprocess_seqid_op */
5495         nfs4_put_stid(&stp->st_stid);
5496 out:
5497         return status;
5498 }
5499
5500 __be32
5501 nfsd4_delegreturn(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5502                   union nfsd4_op_u *u)
5503 {
5504         struct nfsd4_delegreturn *dr = &u->delegreturn;
5505         struct nfs4_delegation *dp;
5506         stateid_t *stateid = &dr->dr_stateid;
5507         struct nfs4_stid *s;
5508         __be32 status;
5509         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
5510
5511         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
5512                 return status;
5513
5514         status = nfsd4_lookup_stateid(cstate, stateid, NFS4_DELEG_STID, &s, nn);
5515         if (status)
5516                 goto out;
5517         dp = delegstateid(s);
5518         status = nfsd4_stid_check_stateid_generation(stateid, &dp->dl_stid, nfsd4_has_session(cstate));
5519         if (status)
5520                 goto put_stateid;
5521
5522         destroy_delegation(dp);
5523 put_stateid:
5524         nfs4_put_stid(&dp->dl_stid);
5525 out:
5526         return status;
5527 }
5528
5529 static inline u64
5530 end_offset(u64 start, u64 len)
5531 {
5532         u64 end;
5533
5534         end = start + len;
5535         return end >= start ? end: NFS4_MAX_UINT64;
5536 }
5537
5538 /* last octet in a range */
5539 static inline u64
5540 last_byte_offset(u64 start, u64 len)
5541 {
5542         u64 end;
5543
5544         WARN_ON_ONCE(!len);
5545         end = start + len;
5546         return end > start ? end - 1: NFS4_MAX_UINT64;
5547 }
5548
5549 /*
5550  * TODO: Linux file offsets are _signed_ 64-bit quantities, which means that
5551  * we can't properly handle lock requests that go beyond the (2^63 - 1)-th
5552  * byte, because of sign extension problems.  Since NFSv4 calls for 64-bit
5553  * locking, this prevents us from being completely protocol-compliant.  The
5554  * real solution to this problem is to start using unsigned file offsets in
5555  * the VFS, but this is a very deep change!
5556  */
5557 static inline void
5558 nfs4_transform_lock_offset(struct file_lock *lock)
5559 {
5560         if (lock->fl_start < 0)
5561                 lock->fl_start = OFFSET_MAX;
5562         if (lock->fl_end < 0)
5563                 lock->fl_end = OFFSET_MAX;
5564 }
5565
5566 static fl_owner_t
5567 nfsd4_fl_get_owner(fl_owner_t owner)
5568 {
5569         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5570
5571         nfs4_get_stateowner(&lo->lo_owner);
5572         return owner;
5573 }
5574
5575 static void
5576 nfsd4_fl_put_owner(fl_owner_t owner)
5577 {
5578         struct nfs4_lockowner *lo = (struct nfs4_lockowner *)owner;
5579
5580         if (lo)
5581                 nfs4_put_stateowner(&lo->lo_owner);
5582 }
5583
5584 static void
5585 nfsd4_lm_notify(struct file_lock *fl)
5586 {
5587         struct nfs4_lockowner           *lo = (struct nfs4_lockowner *)fl->fl_owner;
5588         struct net                      *net = lo->lo_owner.so_client->net;
5589         struct nfsd_net                 *nn = net_generic(net, nfsd_net_id);
5590         struct nfsd4_blocked_lock       *nbl = container_of(fl,
5591                                                 struct nfsd4_blocked_lock, nbl_lock);
5592         bool queue = false;
5593
5594         /* An empty list means that something else is going to be using it */
5595         spin_lock(&nn->blocked_locks_lock);
5596         if (!list_empty(&nbl->nbl_list)) {
5597                 list_del_init(&nbl->nbl_list);
5598                 list_del_init(&nbl->nbl_lru);
5599                 queue = true;
5600         }
5601         spin_unlock(&nn->blocked_locks_lock);
5602
5603         if (queue)
5604                 nfsd4_run_cb(&nbl->nbl_cb);
5605 }
5606
5607 static const struct lock_manager_operations nfsd_posix_mng_ops  = {
5608         .lm_notify = nfsd4_lm_notify,
5609         .lm_get_owner = nfsd4_fl_get_owner,
5610         .lm_put_owner = nfsd4_fl_put_owner,
5611 };
5612
5613 static inline void
5614 nfs4_set_lock_denied(struct file_lock *fl, struct nfsd4_lock_denied *deny)
5615 {
5616         struct nfs4_lockowner *lo;
5617
5618         if (fl->fl_lmops == &nfsd_posix_mng_ops) {
5619                 lo = (struct nfs4_lockowner *) fl->fl_owner;
5620                 deny->ld_owner.data = kmemdup(lo->lo_owner.so_owner.data,
5621                                         lo->lo_owner.so_owner.len, GFP_KERNEL);
5622                 if (!deny->ld_owner.data)
5623                         /* We just don't care that much */
5624                         goto nevermind;
5625                 deny->ld_owner.len = lo->lo_owner.so_owner.len;
5626                 deny->ld_clientid = lo->lo_owner.so_client->cl_clientid;
5627         } else {
5628 nevermind:
5629                 deny->ld_owner.len = 0;
5630                 deny->ld_owner.data = NULL;
5631                 deny->ld_clientid.cl_boot = 0;
5632                 deny->ld_clientid.cl_id = 0;
5633         }
5634         deny->ld_start = fl->fl_start;
5635         deny->ld_length = NFS4_MAX_UINT64;
5636         if (fl->fl_end != NFS4_MAX_UINT64)
5637                 deny->ld_length = fl->fl_end - fl->fl_start + 1;
5638         deny->ld_type = NFS4_READ_LT;
5639         if (fl->fl_type != F_RDLCK)
5640                 deny->ld_type = NFS4_WRITE_LT;
5641 }
5642
5643 static struct nfs4_lockowner *
5644 find_lockowner_str_locked(struct nfs4_client *clp, struct xdr_netobj *owner)
5645 {
5646         unsigned int strhashval = ownerstr_hashval(owner);
5647         struct nfs4_stateowner *so;
5648
5649         lockdep_assert_held(&clp->cl_lock);
5650
5651         list_for_each_entry(so, &clp->cl_ownerstr_hashtbl[strhashval],
5652                             so_strhash) {
5653                 if (so->so_is_open_owner)
5654                         continue;
5655                 if (same_owner_str(so, owner))
5656                         return lockowner(nfs4_get_stateowner(so));
5657         }
5658         return NULL;
5659 }
5660
5661 static struct nfs4_lockowner *
5662 find_lockowner_str(struct nfs4_client *clp, struct xdr_netobj *owner)
5663 {
5664         struct nfs4_lockowner *lo;
5665
5666         spin_lock(&clp->cl_lock);
5667         lo = find_lockowner_str_locked(clp, owner);
5668         spin_unlock(&clp->cl_lock);
5669         return lo;
5670 }
5671
5672 static void nfs4_unhash_lockowner(struct nfs4_stateowner *sop)
5673 {
5674         unhash_lockowner_locked(lockowner(sop));
5675 }
5676
5677 static void nfs4_free_lockowner(struct nfs4_stateowner *sop)
5678 {
5679         struct nfs4_lockowner *lo = lockowner(sop);
5680
5681         kmem_cache_free(lockowner_slab, lo);
5682 }
5683
5684 static const struct nfs4_stateowner_operations lockowner_ops = {
5685         .so_unhash =    nfs4_unhash_lockowner,
5686         .so_free =      nfs4_free_lockowner,
5687 };
5688
5689 /*
5690  * Alloc a lock owner structure.
5691  * Called in nfsd4_lock - therefore, OPEN and OPEN_CONFIRM (if needed) has
5692  * occurred.
5693  *
5694  * strhashval = ownerstr_hashval
5695  */
5696 static struct nfs4_lockowner *
5697 alloc_init_lock_stateowner(unsigned int strhashval, struct nfs4_client *clp,
5698                            struct nfs4_ol_stateid *open_stp,
5699                            struct nfsd4_lock *lock)
5700 {
5701         struct nfs4_lockowner *lo, *ret;
5702
5703         lo = alloc_stateowner(lockowner_slab, &lock->lk_new_owner, clp);
5704         if (!lo)
5705                 return NULL;
5706         INIT_LIST_HEAD(&lo->lo_blocked);
5707         INIT_LIST_HEAD(&lo->lo_owner.so_stateids);
5708         lo->lo_owner.so_is_open_owner = 0;
5709         lo->lo_owner.so_seqid = lock->lk_new_lock_seqid;
5710         lo->lo_owner.so_ops = &lockowner_ops;
5711         spin_lock(&clp->cl_lock);
5712         ret = find_lockowner_str_locked(clp, &lock->lk_new_owner);
5713         if (ret == NULL) {
5714                 list_add(&lo->lo_owner.so_strhash,
5715                          &clp->cl_ownerstr_hashtbl[strhashval]);
5716                 ret = lo;
5717         } else
5718                 nfs4_free_stateowner(&lo->lo_owner);
5719
5720         spin_unlock(&clp->cl_lock);
5721         return ret;
5722 }
5723
5724 static struct nfs4_ol_stateid *
5725 find_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fp)
5726 {
5727         struct nfs4_ol_stateid *lst;
5728         struct nfs4_client *clp = lo->lo_owner.so_client;
5729
5730         lockdep_assert_held(&clp->cl_lock);
5731
5732         list_for_each_entry(lst, &lo->lo_owner.so_stateids, st_perstateowner) {
5733                 if (lst->st_stid.sc_type != NFS4_LOCK_STID)
5734                         continue;
5735                 if (lst->st_stid.sc_file == fp) {
5736                         refcount_inc(&lst->st_stid.sc_count);
5737                         return lst;
5738                 }
5739         }
5740         return NULL;
5741 }
5742
5743 static struct nfs4_ol_stateid *
5744 init_lock_stateid(struct nfs4_ol_stateid *stp, struct nfs4_lockowner *lo,
5745                   struct nfs4_file *fp, struct inode *inode,
5746                   struct nfs4_ol_stateid *open_stp)
5747 {
5748         struct nfs4_client *clp = lo->lo_owner.so_client;
5749         struct nfs4_ol_stateid *retstp;
5750
5751         mutex_init(&stp->st_mutex);
5752         mutex_lock_nested(&stp->st_mutex, OPEN_STATEID_MUTEX);
5753 retry:
5754         spin_lock(&clp->cl_lock);
5755         spin_lock(&fp->fi_lock);
5756         retstp = find_lock_stateid(lo, fp);
5757         if (retstp)
5758                 goto out_unlock;
5759
5760         refcount_inc(&stp->st_stid.sc_count);
5761         stp->st_stid.sc_type = NFS4_LOCK_STID;
5762         stp->st_stateowner = nfs4_get_stateowner(&lo->lo_owner);
5763         get_nfs4_file(fp);
5764         stp->st_stid.sc_file = fp;
5765         stp->st_access_bmap = 0;
5766         stp->st_deny_bmap = open_stp->st_deny_bmap;
5767         stp->st_openstp = open_stp;
5768         list_add(&stp->st_locks, &open_stp->st_locks);
5769         list_add(&stp->st_perstateowner, &lo->lo_owner.so_stateids);
5770         list_add(&stp->st_perfile, &fp->fi_stateids);
5771 out_unlock:
5772         spin_unlock(&fp->fi_lock);
5773         spin_unlock(&clp->cl_lock);
5774         if (retstp) {
5775                 if (nfsd4_lock_ol_stateid(retstp) != nfs_ok) {
5776                         nfs4_put_stid(&retstp->st_stid);
5777                         goto retry;
5778                 }
5779                 /* To keep mutex tracking happy */
5780                 mutex_unlock(&stp->st_mutex);
5781                 stp = retstp;
5782         }
5783         return stp;
5784 }
5785
5786 static struct nfs4_ol_stateid *
5787 find_or_create_lock_stateid(struct nfs4_lockowner *lo, struct nfs4_file *fi,
5788                             struct inode *inode, struct nfs4_ol_stateid *ost,
5789                             bool *new)
5790 {
5791         struct nfs4_stid *ns = NULL;
5792         struct nfs4_ol_stateid *lst;
5793         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5794         struct nfs4_client *clp = oo->oo_owner.so_client;
5795
5796         *new = false;
5797         spin_lock(&clp->cl_lock);
5798         lst = find_lock_stateid(lo, fi);
5799         spin_unlock(&clp->cl_lock);
5800         if (lst != NULL) {
5801                 if (nfsd4_lock_ol_stateid(lst) == nfs_ok)
5802                         goto out;
5803                 nfs4_put_stid(&lst->st_stid);
5804         }
5805         ns = nfs4_alloc_stid(clp, stateid_slab, nfs4_free_lock_stateid);
5806         if (ns == NULL)
5807                 return NULL;
5808
5809         lst = init_lock_stateid(openlockstateid(ns), lo, fi, inode, ost);
5810         if (lst == openlockstateid(ns))
5811                 *new = true;
5812         else
5813                 nfs4_put_stid(ns);
5814 out:
5815         return lst;
5816 }
5817
5818 static int
5819 check_lock_length(u64 offset, u64 length)
5820 {
5821         return ((length == 0) || ((length != NFS4_MAX_UINT64) &&
5822                 (length > ~offset)));
5823 }
5824
5825 static void get_lock_access(struct nfs4_ol_stateid *lock_stp, u32 access)
5826 {
5827         struct nfs4_file *fp = lock_stp->st_stid.sc_file;
5828
5829         lockdep_assert_held(&fp->fi_lock);
5830
5831         if (test_access(access, lock_stp))
5832                 return;
5833         __nfs4_file_get_access(fp, access);
5834         set_access(access, lock_stp);
5835 }
5836
5837 static __be32
5838 lookup_or_create_lock_state(struct nfsd4_compound_state *cstate,
5839                             struct nfs4_ol_stateid *ost,
5840                             struct nfsd4_lock *lock,
5841                             struct nfs4_ol_stateid **plst, bool *new)
5842 {
5843         __be32 status;
5844         struct nfs4_file *fi = ost->st_stid.sc_file;
5845         struct nfs4_openowner *oo = openowner(ost->st_stateowner);
5846         struct nfs4_client *cl = oo->oo_owner.so_client;
5847         struct inode *inode = d_inode(cstate->current_fh.fh_dentry);
5848         struct nfs4_lockowner *lo;
5849         struct nfs4_ol_stateid *lst;
5850         unsigned int strhashval;
5851
5852         lo = find_lockowner_str(cl, &lock->lk_new_owner);
5853         if (!lo) {
5854                 strhashval = ownerstr_hashval(&lock->lk_new_owner);
5855                 lo = alloc_init_lock_stateowner(strhashval, cl, ost, lock);
5856                 if (lo == NULL)
5857                         return nfserr_jukebox;
5858         } else {
5859                 /* with an existing lockowner, seqids must be the same */
5860                 status = nfserr_bad_seqid;
5861                 if (!cstate->minorversion &&
5862                     lock->lk_new_lock_seqid != lo->lo_owner.so_seqid)
5863                         goto out;
5864         }
5865
5866         lst = find_or_create_lock_stateid(lo, fi, inode, ost, new);
5867         if (lst == NULL) {
5868                 status = nfserr_jukebox;
5869                 goto out;
5870         }
5871
5872         status = nfs_ok;
5873         *plst = lst;
5874 out:
5875         nfs4_put_stateowner(&lo->lo_owner);
5876         return status;
5877 }
5878
5879 /*
5880  *  LOCK operation
5881  */
5882 __be32
5883 nfsd4_lock(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
5884            union nfsd4_op_u *u)
5885 {
5886         struct nfsd4_lock *lock = &u->lock;
5887         struct nfs4_openowner *open_sop = NULL;
5888         struct nfs4_lockowner *lock_sop = NULL;
5889         struct nfs4_ol_stateid *lock_stp = NULL;
5890         struct nfs4_ol_stateid *open_stp = NULL;
5891         struct nfs4_file *fp;
5892         struct file *filp = NULL;
5893         struct nfsd4_blocked_lock *nbl = NULL;
5894         struct file_lock *file_lock = NULL;
5895         struct file_lock *conflock = NULL;
5896         __be32 status = 0;
5897         int lkflg;
5898         int err;
5899         bool new = false;
5900         unsigned char fl_type;
5901         unsigned int fl_flags = FL_POSIX;
5902         struct net *net = SVC_NET(rqstp);
5903         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
5904
5905         dprintk("NFSD: nfsd4_lock: start=%Ld length=%Ld\n",
5906                 (long long) lock->lk_offset,
5907                 (long long) lock->lk_length);
5908
5909         if (check_lock_length(lock->lk_offset, lock->lk_length))
5910                  return nfserr_inval;
5911
5912         if ((status = fh_verify(rqstp, &cstate->current_fh,
5913                                 S_IFREG, NFSD_MAY_LOCK))) {
5914                 dprintk("NFSD: nfsd4_lock: permission denied!\n");
5915                 return status;
5916         }
5917
5918         if (lock->lk_is_new) {
5919                 if (nfsd4_has_session(cstate))
5920                         /* See rfc 5661 18.10.3: given clientid is ignored: */
5921                         memcpy(&lock->lk_new_clientid,
5922                                 &cstate->session->se_client->cl_clientid,
5923                                 sizeof(clientid_t));
5924
5925                 status = nfserr_stale_clientid;
5926                 if (STALE_CLIENTID(&lock->lk_new_clientid, nn))
5927                         goto out;
5928
5929                 /* validate and update open stateid and open seqid */
5930                 status = nfs4_preprocess_confirmed_seqid_op(cstate,
5931                                         lock->lk_new_open_seqid,
5932                                         &lock->lk_new_open_stateid,
5933                                         &open_stp, nn);
5934                 if (status)
5935                         goto out;
5936                 mutex_unlock(&open_stp->st_mutex);
5937                 open_sop = openowner(open_stp->st_stateowner);
5938                 status = nfserr_bad_stateid;
5939                 if (!same_clid(&open_sop->oo_owner.so_client->cl_clientid,
5940                                                 &lock->lk_new_clientid))
5941                         goto out;
5942                 status = lookup_or_create_lock_state(cstate, open_stp, lock,
5943                                                         &lock_stp, &new);
5944         } else {
5945                 status = nfs4_preprocess_seqid_op(cstate,
5946                                        lock->lk_old_lock_seqid,
5947                                        &lock->lk_old_lock_stateid,
5948                                        NFS4_LOCK_STID, &lock_stp, nn);
5949         }
5950         if (status)
5951                 goto out;
5952         lock_sop = lockowner(lock_stp->st_stateowner);
5953
5954         lkflg = setlkflg(lock->lk_type);
5955         status = nfs4_check_openmode(lock_stp, lkflg);
5956         if (status)
5957                 goto out;
5958
5959         status = nfserr_grace;
5960         if (locks_in_grace(net) && !lock->lk_reclaim)
5961                 goto out;
5962         status = nfserr_no_grace;
5963         if (!locks_in_grace(net) && lock->lk_reclaim)
5964                 goto out;
5965
5966         fp = lock_stp->st_stid.sc_file;
5967         switch (lock->lk_type) {
5968                 case NFS4_READW_LT:
5969                         if (nfsd4_has_session(cstate))
5970                                 fl_flags |= FL_SLEEP;
5971                         /* Fallthrough */
5972                 case NFS4_READ_LT:
5973                         spin_lock(&fp->fi_lock);
5974                         filp = find_readable_file_locked(fp);
5975                         if (filp)
5976                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_READ);
5977                         spin_unlock(&fp->fi_lock);
5978                         fl_type = F_RDLCK;
5979                         break;
5980                 case NFS4_WRITEW_LT:
5981                         if (nfsd4_has_session(cstate))
5982                                 fl_flags |= FL_SLEEP;
5983                         /* Fallthrough */
5984                 case NFS4_WRITE_LT:
5985                         spin_lock(&fp->fi_lock);
5986                         filp = find_writeable_file_locked(fp);
5987                         if (filp)
5988                                 get_lock_access(lock_stp, NFS4_SHARE_ACCESS_WRITE);
5989                         spin_unlock(&fp->fi_lock);
5990                         fl_type = F_WRLCK;
5991                         break;
5992                 default:
5993                         status = nfserr_inval;
5994                 goto out;
5995         }
5996
5997         if (!filp) {
5998                 status = nfserr_openmode;
5999                 goto out;
6000         }
6001
6002         nbl = find_or_allocate_block(lock_sop, &fp->fi_fhandle, nn);
6003         if (!nbl) {
6004                 dprintk("NFSD: %s: unable to allocate block!\n", __func__);
6005                 status = nfserr_jukebox;
6006                 goto out;
6007         }
6008
6009         file_lock = &nbl->nbl_lock;
6010         file_lock->fl_type = fl_type;
6011         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(&lock_sop->lo_owner));
6012         file_lock->fl_pid = current->tgid;
6013         file_lock->fl_file = filp;
6014         file_lock->fl_flags = fl_flags;
6015         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6016         file_lock->fl_start = lock->lk_offset;
6017         file_lock->fl_end = last_byte_offset(lock->lk_offset, lock->lk_length);
6018         nfs4_transform_lock_offset(file_lock);
6019
6020         conflock = locks_alloc_lock();
6021         if (!conflock) {
6022                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6023                 status = nfserr_jukebox;
6024                 goto out;
6025         }
6026
6027         if (fl_flags & FL_SLEEP) {
6028                 nbl->nbl_time = jiffies;
6029                 spin_lock(&nn->blocked_locks_lock);
6030                 list_add_tail(&nbl->nbl_list, &lock_sop->lo_blocked);
6031                 list_add_tail(&nbl->nbl_lru, &nn->blocked_locks_lru);
6032                 spin_unlock(&nn->blocked_locks_lock);
6033         }
6034
6035         err = vfs_lock_file(filp, F_SETLK, file_lock, conflock);
6036         switch (err) {
6037         case 0: /* success! */
6038                 nfs4_inc_and_copy_stateid(&lock->lk_resp_stateid, &lock_stp->st_stid);
6039                 status = 0;
6040                 break;
6041         case FILE_LOCK_DEFERRED:
6042                 nbl = NULL;
6043                 /* Fallthrough */
6044         case -EAGAIN:           /* conflock holds conflicting lock */
6045                 status = nfserr_denied;
6046                 dprintk("NFSD: nfsd4_lock: conflicting lock found!\n");
6047                 nfs4_set_lock_denied(conflock, &lock->lk_denied);
6048                 break;
6049         case -EDEADLK:
6050                 status = nfserr_deadlock;
6051                 break;
6052         default:
6053                 dprintk("NFSD: nfsd4_lock: vfs_lock_file() failed! status %d\n",err);
6054                 status = nfserrno(err);
6055                 break;
6056         }
6057 out:
6058         if (nbl) {
6059                 /* dequeue it if we queued it before */
6060                 if (fl_flags & FL_SLEEP) {
6061                         spin_lock(&nn->blocked_locks_lock);
6062                         list_del_init(&nbl->nbl_list);
6063                         list_del_init(&nbl->nbl_lru);
6064                         spin_unlock(&nn->blocked_locks_lock);
6065                 }
6066                 free_blocked_lock(nbl);
6067         }
6068         if (filp)
6069                 fput(filp);
6070         if (lock_stp) {
6071                 /* Bump seqid manually if the 4.0 replay owner is openowner */
6072                 if (cstate->replay_owner &&
6073                     cstate->replay_owner != &lock_sop->lo_owner &&
6074                     seqid_mutating_err(ntohl(status)))
6075                         lock_sop->lo_owner.so_seqid++;
6076
6077                 /*
6078                  * If this is a new, never-before-used stateid, and we are
6079                  * returning an error, then just go ahead and release it.
6080                  */
6081                 if (status && new) {
6082                         lock_stp->st_stid.sc_type = NFS4_CLOSED_STID;
6083                         release_lock_stateid(lock_stp);
6084                 }
6085
6086                 mutex_unlock(&lock_stp->st_mutex);
6087
6088                 nfs4_put_stid(&lock_stp->st_stid);
6089         }
6090         if (open_stp)
6091                 nfs4_put_stid(&open_stp->st_stid);
6092         nfsd4_bump_seqid(cstate, status);
6093         if (conflock)
6094                 locks_free_lock(conflock);
6095         return status;
6096 }
6097
6098 /*
6099  * The NFSv4 spec allows a client to do a LOCKT without holding an OPEN,
6100  * so we do a temporary open here just to get an open file to pass to
6101  * vfs_test_lock.  (Arguably perhaps test_lock should be done with an
6102  * inode operation.)
6103  */
6104 static __be32 nfsd_test_lock(struct svc_rqst *rqstp, struct svc_fh *fhp, struct file_lock *lock)
6105 {
6106         struct file *file;
6107         __be32 err = nfsd_open(rqstp, fhp, S_IFREG, NFSD_MAY_READ, &file);
6108         if (!err) {
6109                 err = nfserrno(vfs_test_lock(file, lock));
6110                 fput(file);
6111         }
6112         return err;
6113 }
6114
6115 /*
6116  * LOCKT operation
6117  */
6118 __be32
6119 nfsd4_lockt(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6120             union nfsd4_op_u *u)
6121 {
6122         struct nfsd4_lockt *lockt = &u->lockt;
6123         struct file_lock *file_lock = NULL;
6124         struct nfs4_lockowner *lo = NULL;
6125         __be32 status;
6126         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6127
6128         if (locks_in_grace(SVC_NET(rqstp)))
6129                 return nfserr_grace;
6130
6131         if (check_lock_length(lockt->lt_offset, lockt->lt_length))
6132                  return nfserr_inval;
6133
6134         if (!nfsd4_has_session(cstate)) {
6135                 status = lookup_clientid(&lockt->lt_clientid, cstate, nn);
6136                 if (status)
6137                         goto out;
6138         }
6139
6140         if ((status = fh_verify(rqstp, &cstate->current_fh, S_IFREG, 0)))
6141                 goto out;
6142
6143         file_lock = locks_alloc_lock();
6144         if (!file_lock) {
6145                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6146                 status = nfserr_jukebox;
6147                 goto out;
6148         }
6149
6150         switch (lockt->lt_type) {
6151                 case NFS4_READ_LT:
6152                 case NFS4_READW_LT:
6153                         file_lock->fl_type = F_RDLCK;
6154                 break;
6155                 case NFS4_WRITE_LT:
6156                 case NFS4_WRITEW_LT:
6157                         file_lock->fl_type = F_WRLCK;
6158                 break;
6159                 default:
6160                         dprintk("NFSD: nfs4_lockt: bad lock type!\n");
6161                         status = nfserr_inval;
6162                 goto out;
6163         }
6164
6165         lo = find_lockowner_str(cstate->clp, &lockt->lt_owner);
6166         if (lo)
6167                 file_lock->fl_owner = (fl_owner_t)lo;
6168         file_lock->fl_pid = current->tgid;
6169         file_lock->fl_flags = FL_POSIX;
6170
6171         file_lock->fl_start = lockt->lt_offset;
6172         file_lock->fl_end = last_byte_offset(lockt->lt_offset, lockt->lt_length);
6173
6174         nfs4_transform_lock_offset(file_lock);
6175
6176         status = nfsd_test_lock(rqstp, &cstate->current_fh, file_lock);
6177         if (status)
6178                 goto out;
6179
6180         if (file_lock->fl_type != F_UNLCK) {
6181                 status = nfserr_denied;
6182                 nfs4_set_lock_denied(file_lock, &lockt->lt_denied);
6183         }
6184 out:
6185         if (lo)
6186                 nfs4_put_stateowner(&lo->lo_owner);
6187         if (file_lock)
6188                 locks_free_lock(file_lock);
6189         return status;
6190 }
6191
6192 __be32
6193 nfsd4_locku(struct svc_rqst *rqstp, struct nfsd4_compound_state *cstate,
6194             union nfsd4_op_u *u)
6195 {
6196         struct nfsd4_locku *locku = &u->locku;
6197         struct nfs4_ol_stateid *stp;
6198         struct file *filp = NULL;
6199         struct file_lock *file_lock = NULL;
6200         __be32 status;
6201         int err;
6202         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6203
6204         dprintk("NFSD: nfsd4_locku: start=%Ld length=%Ld\n",
6205                 (long long) locku->lu_offset,
6206                 (long long) locku->lu_length);
6207
6208         if (check_lock_length(locku->lu_offset, locku->lu_length))
6209                  return nfserr_inval;
6210
6211         status = nfs4_preprocess_seqid_op(cstate, locku->lu_seqid,
6212                                         &locku->lu_stateid, NFS4_LOCK_STID,
6213                                         &stp, nn);
6214         if (status)
6215                 goto out;
6216         filp = find_any_file(stp->st_stid.sc_file);
6217         if (!filp) {
6218                 status = nfserr_lock_range;
6219                 goto put_stateid;
6220         }
6221         file_lock = locks_alloc_lock();
6222         if (!file_lock) {
6223                 dprintk("NFSD: %s: unable to allocate lock!\n", __func__);
6224                 status = nfserr_jukebox;
6225                 goto fput;
6226         }
6227
6228         file_lock->fl_type = F_UNLCK;
6229         file_lock->fl_owner = (fl_owner_t)lockowner(nfs4_get_stateowner(stp->st_stateowner));
6230         file_lock->fl_pid = current->tgid;
6231         file_lock->fl_file = filp;
6232         file_lock->fl_flags = FL_POSIX;
6233         file_lock->fl_lmops = &nfsd_posix_mng_ops;
6234         file_lock->fl_start = locku->lu_offset;
6235
6236         file_lock->fl_end = last_byte_offset(locku->lu_offset,
6237                                                 locku->lu_length);
6238         nfs4_transform_lock_offset(file_lock);
6239
6240         err = vfs_lock_file(filp, F_SETLK, file_lock, NULL);
6241         if (err) {
6242                 dprintk("NFSD: nfs4_locku: vfs_lock_file failed!\n");
6243                 goto out_nfserr;
6244         }
6245         nfs4_inc_and_copy_stateid(&locku->lu_stateid, &stp->st_stid);
6246 fput:
6247         fput(filp);
6248 put_stateid:
6249         mutex_unlock(&stp->st_mutex);
6250         nfs4_put_stid(&stp->st_stid);
6251 out:
6252         nfsd4_bump_seqid(cstate, status);
6253         if (file_lock)
6254                 locks_free_lock(file_lock);
6255         return status;
6256
6257 out_nfserr:
6258         status = nfserrno(err);
6259         goto fput;
6260 }
6261
6262 /*
6263  * returns
6264  *      true:  locks held by lockowner
6265  *      false: no locks held by lockowner
6266  */
6267 static bool
6268 check_for_locks(struct nfs4_file *fp, struct nfs4_lockowner *lowner)
6269 {
6270         struct file_lock *fl;
6271         int status = false;
6272         struct file *filp = find_any_file(fp);
6273         struct inode *inode;
6274         struct file_lock_context *flctx;
6275
6276         if (!filp) {
6277                 /* Any valid lock stateid should have some sort of access */
6278                 WARN_ON_ONCE(1);
6279                 return status;
6280         }
6281
6282         inode = file_inode(filp);
6283         flctx = inode->i_flctx;
6284
6285         if (flctx && !list_empty_careful(&flctx->flc_posix)) {
6286                 spin_lock(&flctx->flc_lock);
6287                 list_for_each_entry(fl, &flctx->flc_posix, fl_list) {
6288                         if (fl->fl_owner == (fl_owner_t)lowner) {
6289                                 status = true;
6290                                 break;
6291                         }
6292                 }
6293                 spin_unlock(&flctx->flc_lock);
6294         }
6295         fput(filp);
6296         return status;
6297 }
6298
6299 __be32
6300 nfsd4_release_lockowner(struct svc_rqst *rqstp,
6301                         struct nfsd4_compound_state *cstate,
6302                         union nfsd4_op_u *u)
6303 {
6304         struct nfsd4_release_lockowner *rlockowner = &u->release_lockowner;
6305         clientid_t *clid = &rlockowner->rl_clientid;
6306         struct nfs4_stateowner *sop;
6307         struct nfs4_lockowner *lo = NULL;
6308         struct nfs4_ol_stateid *stp;
6309         struct xdr_netobj *owner = &rlockowner->rl_owner;
6310         unsigned int hashval = ownerstr_hashval(owner);
6311         __be32 status;
6312         struct nfsd_net *nn = net_generic(SVC_NET(rqstp), nfsd_net_id);
6313         struct nfs4_client *clp;
6314         LIST_HEAD (reaplist);
6315
6316         dprintk("nfsd4_release_lockowner clientid: (%08x/%08x):\n",
6317                 clid->cl_boot, clid->cl_id);
6318
6319         status = lookup_clientid(clid, cstate, nn);
6320         if (status)
6321                 return status;
6322
6323         clp = cstate->clp;
6324         /* Find the matching lock stateowner */
6325         spin_lock(&clp->cl_lock);
6326         list_for_each_entry(sop, &clp->cl_ownerstr_hashtbl[hashval],
6327                             so_strhash) {
6328
6329                 if (sop->so_is_open_owner || !same_owner_str(sop, owner))
6330                         continue;
6331
6332                 /* see if there are still any locks associated with it */
6333                 lo = lockowner(sop);
6334                 list_for_each_entry(stp, &sop->so_stateids, st_perstateowner) {
6335                         if (check_for_locks(stp->st_stid.sc_file, lo)) {
6336                                 status = nfserr_locks_held;
6337                                 spin_unlock(&clp->cl_lock);
6338                                 return status;
6339                         }
6340                 }
6341
6342                 nfs4_get_stateowner(sop);
6343                 break;
6344         }
6345         if (!lo) {
6346                 spin_unlock(&clp->cl_lock);
6347                 return status;
6348         }
6349
6350         unhash_lockowner_locked(lo);
6351         while (!list_empty(&lo->lo_owner.so_stateids)) {
6352                 stp = list_first_entry(&lo->lo_owner.so_stateids,
6353                                        struct nfs4_ol_stateid,
6354                                        st_perstateowner);
6355                 WARN_ON(!unhash_lock_stateid(stp));
6356                 put_ol_stateid_locked(stp, &reaplist);
6357         }
6358         spin_unlock(&clp->cl_lock);
6359         free_ol_stateid_reaplist(&reaplist);
6360         nfs4_put_stateowner(&lo->lo_owner);
6361
6362         return status;
6363 }
6364
6365 static inline struct nfs4_client_reclaim *
6366 alloc_reclaim(void)
6367 {
6368         return kmalloc(sizeof(struct nfs4_client_reclaim), GFP_KERNEL);
6369 }
6370
6371 bool
6372 nfs4_has_reclaimed_state(const char *name, struct nfsd_net *nn)
6373 {
6374         struct nfs4_client_reclaim *crp;
6375
6376         crp = nfsd4_find_reclaim_client(name, nn);
6377         return (crp && crp->cr_clp);
6378 }
6379
6380 /*
6381  * failure => all reset bets are off, nfserr_no_grace...
6382  */
6383 struct nfs4_client_reclaim *
6384 nfs4_client_to_reclaim(const char *name, struct nfsd_net *nn)
6385 {
6386         unsigned int strhashval;
6387         struct nfs4_client_reclaim *crp;
6388
6389         dprintk("NFSD nfs4_client_to_reclaim NAME: %.*s\n", HEXDIR_LEN, name);
6390         crp = alloc_reclaim();
6391         if (crp) {
6392                 strhashval = clientstr_hashval(name);
6393                 INIT_LIST_HEAD(&crp->cr_strhash);
6394                 list_add(&crp->cr_strhash, &nn->reclaim_str_hashtbl[strhashval]);
6395                 memcpy(crp->cr_recdir, name, HEXDIR_LEN);
6396                 crp->cr_clp = NULL;
6397                 nn->reclaim_str_hashtbl_size++;
6398         }
6399         return crp;
6400 }
6401
6402 void
6403 nfs4_remove_reclaim_record(struct nfs4_client_reclaim *crp, struct nfsd_net *nn)
6404 {
6405         list_del(&crp->cr_strhash);
6406         kfree(crp);
6407         nn->reclaim_str_hashtbl_size--;
6408 }
6409
6410 void
6411 nfs4_release_reclaim(struct nfsd_net *nn)
6412 {
6413         struct nfs4_client_reclaim *crp = NULL;
6414         int i;
6415
6416         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
6417                 while (!list_empty(&nn->reclaim_str_hashtbl[i])) {
6418                         crp = list_entry(nn->reclaim_str_hashtbl[i].next,
6419                                         struct nfs4_client_reclaim, cr_strhash);
6420                         nfs4_remove_reclaim_record(crp, nn);
6421                 }
6422         }
6423         WARN_ON_ONCE(nn->reclaim_str_hashtbl_size);
6424 }
6425
6426 /*
6427  * called from OPEN, CLAIM_PREVIOUS with a new clientid. */
6428 struct nfs4_client_reclaim *
6429 nfsd4_find_reclaim_client(const char *recdir, struct nfsd_net *nn)
6430 {
6431         unsigned int strhashval;
6432         struct nfs4_client_reclaim *crp = NULL;
6433
6434         dprintk("NFSD: nfs4_find_reclaim_client for recdir %s\n", recdir);
6435
6436         strhashval = clientstr_hashval(recdir);
6437         list_for_each_entry(crp, &nn->reclaim_str_hashtbl[strhashval], cr_strhash) {
6438                 if (same_name(crp->cr_recdir, recdir)) {
6439                         return crp;
6440                 }
6441         }
6442         return NULL;
6443 }
6444
6445 /*
6446 * Called from OPEN. Look for clientid in reclaim list.
6447 */
6448 __be32
6449 nfs4_check_open_reclaim(clientid_t *clid,
6450                 struct nfsd4_compound_state *cstate,
6451                 struct nfsd_net *nn)
6452 {
6453         __be32 status;
6454
6455         /* find clientid in conf_id_hashtbl */
6456         status = lookup_clientid(clid, cstate, nn);
6457         if (status)
6458                 return nfserr_reclaim_bad;
6459
6460         if (test_bit(NFSD4_CLIENT_RECLAIM_COMPLETE, &cstate->clp->cl_flags))
6461                 return nfserr_no_grace;
6462
6463         if (nfsd4_client_record_check(cstate->clp))
6464                 return nfserr_reclaim_bad;
6465
6466         return nfs_ok;
6467 }
6468
6469 #ifdef CONFIG_NFSD_FAULT_INJECTION
6470 static inline void
6471 put_client(struct nfs4_client *clp)
6472 {
6473         atomic_dec(&clp->cl_refcount);
6474 }
6475
6476 static struct nfs4_client *
6477 nfsd_find_client(struct sockaddr_storage *addr, size_t addr_size)
6478 {
6479         struct nfs4_client *clp;
6480         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6481                                           nfsd_net_id);
6482
6483         if (!nfsd_netns_ready(nn))
6484                 return NULL;
6485
6486         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6487                 if (memcmp(&clp->cl_addr, addr, addr_size) == 0)
6488                         return clp;
6489         }
6490         return NULL;
6491 }
6492
6493 u64
6494 nfsd_inject_print_clients(void)
6495 {
6496         struct nfs4_client *clp;
6497         u64 count = 0;
6498         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6499                                           nfsd_net_id);
6500         char buf[INET6_ADDRSTRLEN];
6501
6502         if (!nfsd_netns_ready(nn))
6503                 return 0;
6504
6505         spin_lock(&nn->client_lock);
6506         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6507                 rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6508                 pr_info("NFS Client: %s\n", buf);
6509                 ++count;
6510         }
6511         spin_unlock(&nn->client_lock);
6512
6513         return count;
6514 }
6515
6516 u64
6517 nfsd_inject_forget_client(struct sockaddr_storage *addr, size_t addr_size)
6518 {
6519         u64 count = 0;
6520         struct nfs4_client *clp;
6521         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6522                                           nfsd_net_id);
6523
6524         if (!nfsd_netns_ready(nn))
6525                 return count;
6526
6527         spin_lock(&nn->client_lock);
6528         clp = nfsd_find_client(addr, addr_size);
6529         if (clp) {
6530                 if (mark_client_expired_locked(clp) == nfs_ok)
6531                         ++count;
6532                 else
6533                         clp = NULL;
6534         }
6535         spin_unlock(&nn->client_lock);
6536
6537         if (clp)
6538                 expire_client(clp);
6539
6540         return count;
6541 }
6542
6543 u64
6544 nfsd_inject_forget_clients(u64 max)
6545 {
6546         u64 count = 0;
6547         struct nfs4_client *clp, *next;
6548         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6549                                                 nfsd_net_id);
6550         LIST_HEAD(reaplist);
6551
6552         if (!nfsd_netns_ready(nn))
6553                 return count;
6554
6555         spin_lock(&nn->client_lock);
6556         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
6557                 if (mark_client_expired_locked(clp) == nfs_ok) {
6558                         list_add(&clp->cl_lru, &reaplist);
6559                         if (max != 0 && ++count >= max)
6560                                 break;
6561                 }
6562         }
6563         spin_unlock(&nn->client_lock);
6564
6565         list_for_each_entry_safe(clp, next, &reaplist, cl_lru)
6566                 expire_client(clp);
6567
6568         return count;
6569 }
6570
6571 static void nfsd_print_count(struct nfs4_client *clp, unsigned int count,
6572                              const char *type)
6573 {
6574         char buf[INET6_ADDRSTRLEN];
6575         rpc_ntop((struct sockaddr *)&clp->cl_addr, buf, sizeof(buf));
6576         printk(KERN_INFO "NFS Client: %s has %u %s\n", buf, count, type);
6577 }
6578
6579 static void
6580 nfsd_inject_add_lock_to_list(struct nfs4_ol_stateid *lst,
6581                              struct list_head *collect)
6582 {
6583         struct nfs4_client *clp = lst->st_stid.sc_client;
6584         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6585                                           nfsd_net_id);
6586
6587         if (!collect)
6588                 return;
6589
6590         lockdep_assert_held(&nn->client_lock);
6591         atomic_inc(&clp->cl_refcount);
6592         list_add(&lst->st_locks, collect);
6593 }
6594
6595 static u64 nfsd_foreach_client_lock(struct nfs4_client *clp, u64 max,
6596                                     struct list_head *collect,
6597                                     bool (*func)(struct nfs4_ol_stateid *))
6598 {
6599         struct nfs4_openowner *oop;
6600         struct nfs4_ol_stateid *stp, *st_next;
6601         struct nfs4_ol_stateid *lst, *lst_next;
6602         u64 count = 0;
6603
6604         spin_lock(&clp->cl_lock);
6605         list_for_each_entry(oop, &clp->cl_openowners, oo_perclient) {
6606                 list_for_each_entry_safe(stp, st_next,
6607                                 &oop->oo_owner.so_stateids, st_perstateowner) {
6608                         list_for_each_entry_safe(lst, lst_next,
6609                                         &stp->st_locks, st_locks) {
6610                                 if (func) {
6611                                         if (func(lst))
6612                                                 nfsd_inject_add_lock_to_list(lst,
6613                                                                         collect);
6614                                 }
6615                                 ++count;
6616                                 /*
6617                                  * Despite the fact that these functions deal
6618                                  * with 64-bit integers for "count", we must
6619                                  * ensure that it doesn't blow up the
6620                                  * clp->cl_refcount. Throw a warning if we
6621                                  * start to approach INT_MAX here.
6622                                  */
6623                                 WARN_ON_ONCE(count == (INT_MAX / 2));
6624                                 if (count == max)
6625                                         goto out;
6626                         }
6627                 }
6628         }
6629 out:
6630         spin_unlock(&clp->cl_lock);
6631
6632         return count;
6633 }
6634
6635 static u64
6636 nfsd_collect_client_locks(struct nfs4_client *clp, struct list_head *collect,
6637                           u64 max)
6638 {
6639         return nfsd_foreach_client_lock(clp, max, collect, unhash_lock_stateid);
6640 }
6641
6642 static u64
6643 nfsd_print_client_locks(struct nfs4_client *clp)
6644 {
6645         u64 count = nfsd_foreach_client_lock(clp, 0, NULL, NULL);
6646         nfsd_print_count(clp, count, "locked files");
6647         return count;
6648 }
6649
6650 u64
6651 nfsd_inject_print_locks(void)
6652 {
6653         struct nfs4_client *clp;
6654         u64 count = 0;
6655         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6656                                                 nfsd_net_id);
6657
6658         if (!nfsd_netns_ready(nn))
6659                 return 0;
6660
6661         spin_lock(&nn->client_lock);
6662         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6663                 count += nfsd_print_client_locks(clp);
6664         spin_unlock(&nn->client_lock);
6665
6666         return count;
6667 }
6668
6669 static void
6670 nfsd_reap_locks(struct list_head *reaplist)
6671 {
6672         struct nfs4_client *clp;
6673         struct nfs4_ol_stateid *stp, *next;
6674
6675         list_for_each_entry_safe(stp, next, reaplist, st_locks) {
6676                 list_del_init(&stp->st_locks);
6677                 clp = stp->st_stid.sc_client;
6678                 nfs4_put_stid(&stp->st_stid);
6679                 put_client(clp);
6680         }
6681 }
6682
6683 u64
6684 nfsd_inject_forget_client_locks(struct sockaddr_storage *addr, size_t addr_size)
6685 {
6686         unsigned int count = 0;
6687         struct nfs4_client *clp;
6688         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6689                                                 nfsd_net_id);
6690         LIST_HEAD(reaplist);
6691
6692         if (!nfsd_netns_ready(nn))
6693                 return count;
6694
6695         spin_lock(&nn->client_lock);
6696         clp = nfsd_find_client(addr, addr_size);
6697         if (clp)
6698                 count = nfsd_collect_client_locks(clp, &reaplist, 0);
6699         spin_unlock(&nn->client_lock);
6700         nfsd_reap_locks(&reaplist);
6701         return count;
6702 }
6703
6704 u64
6705 nfsd_inject_forget_locks(u64 max)
6706 {
6707         u64 count = 0;
6708         struct nfs4_client *clp;
6709         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6710                                                 nfsd_net_id);
6711         LIST_HEAD(reaplist);
6712
6713         if (!nfsd_netns_ready(nn))
6714                 return count;
6715
6716         spin_lock(&nn->client_lock);
6717         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6718                 count += nfsd_collect_client_locks(clp, &reaplist, max - count);
6719                 if (max != 0 && count >= max)
6720                         break;
6721         }
6722         spin_unlock(&nn->client_lock);
6723         nfsd_reap_locks(&reaplist);
6724         return count;
6725 }
6726
6727 static u64
6728 nfsd_foreach_client_openowner(struct nfs4_client *clp, u64 max,
6729                               struct list_head *collect,
6730                               void (*func)(struct nfs4_openowner *))
6731 {
6732         struct nfs4_openowner *oop, *next;
6733         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6734                                                 nfsd_net_id);
6735         u64 count = 0;
6736
6737         lockdep_assert_held(&nn->client_lock);
6738
6739         spin_lock(&clp->cl_lock);
6740         list_for_each_entry_safe(oop, next, &clp->cl_openowners, oo_perclient) {
6741                 if (func) {
6742                         func(oop);
6743                         if (collect) {
6744                                 atomic_inc(&clp->cl_refcount);
6745                                 list_add(&oop->oo_perclient, collect);
6746                         }
6747                 }
6748                 ++count;
6749                 /*
6750                  * Despite the fact that these functions deal with
6751                  * 64-bit integers for "count", we must ensure that
6752                  * it doesn't blow up the clp->cl_refcount. Throw a
6753                  * warning if we start to approach INT_MAX here.
6754                  */
6755                 WARN_ON_ONCE(count == (INT_MAX / 2));
6756                 if (count == max)
6757                         break;
6758         }
6759         spin_unlock(&clp->cl_lock);
6760
6761         return count;
6762 }
6763
6764 static u64
6765 nfsd_print_client_openowners(struct nfs4_client *clp)
6766 {
6767         u64 count = nfsd_foreach_client_openowner(clp, 0, NULL, NULL);
6768
6769         nfsd_print_count(clp, count, "openowners");
6770         return count;
6771 }
6772
6773 static u64
6774 nfsd_collect_client_openowners(struct nfs4_client *clp,
6775                                struct list_head *collect, u64 max)
6776 {
6777         return nfsd_foreach_client_openowner(clp, max, collect,
6778                                                 unhash_openowner_locked);
6779 }
6780
6781 u64
6782 nfsd_inject_print_openowners(void)
6783 {
6784         struct nfs4_client *clp;
6785         u64 count = 0;
6786         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6787                                                 nfsd_net_id);
6788
6789         if (!nfsd_netns_ready(nn))
6790                 return 0;
6791
6792         spin_lock(&nn->client_lock);
6793         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6794                 count += nfsd_print_client_openowners(clp);
6795         spin_unlock(&nn->client_lock);
6796
6797         return count;
6798 }
6799
6800 static void
6801 nfsd_reap_openowners(struct list_head *reaplist)
6802 {
6803         struct nfs4_client *clp;
6804         struct nfs4_openowner *oop, *next;
6805
6806         list_for_each_entry_safe(oop, next, reaplist, oo_perclient) {
6807                 list_del_init(&oop->oo_perclient);
6808                 clp = oop->oo_owner.so_client;
6809                 release_openowner(oop);
6810                 put_client(clp);
6811         }
6812 }
6813
6814 u64
6815 nfsd_inject_forget_client_openowners(struct sockaddr_storage *addr,
6816                                      size_t addr_size)
6817 {
6818         unsigned int count = 0;
6819         struct nfs4_client *clp;
6820         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6821                                                 nfsd_net_id);
6822         LIST_HEAD(reaplist);
6823
6824         if (!nfsd_netns_ready(nn))
6825                 return count;
6826
6827         spin_lock(&nn->client_lock);
6828         clp = nfsd_find_client(addr, addr_size);
6829         if (clp)
6830                 count = nfsd_collect_client_openowners(clp, &reaplist, 0);
6831         spin_unlock(&nn->client_lock);
6832         nfsd_reap_openowners(&reaplist);
6833         return count;
6834 }
6835
6836 u64
6837 nfsd_inject_forget_openowners(u64 max)
6838 {
6839         u64 count = 0;
6840         struct nfs4_client *clp;
6841         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6842                                                 nfsd_net_id);
6843         LIST_HEAD(reaplist);
6844
6845         if (!nfsd_netns_ready(nn))
6846                 return count;
6847
6848         spin_lock(&nn->client_lock);
6849         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6850                 count += nfsd_collect_client_openowners(clp, &reaplist,
6851                                                         max - count);
6852                 if (max != 0 && count >= max)
6853                         break;
6854         }
6855         spin_unlock(&nn->client_lock);
6856         nfsd_reap_openowners(&reaplist);
6857         return count;
6858 }
6859
6860 static u64 nfsd_find_all_delegations(struct nfs4_client *clp, u64 max,
6861                                      struct list_head *victims)
6862 {
6863         struct nfs4_delegation *dp, *next;
6864         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6865                                                 nfsd_net_id);
6866         u64 count = 0;
6867
6868         lockdep_assert_held(&nn->client_lock);
6869
6870         spin_lock(&state_lock);
6871         list_for_each_entry_safe(dp, next, &clp->cl_delegations, dl_perclnt) {
6872                 if (victims) {
6873                         /*
6874                          * It's not safe to mess with delegations that have a
6875                          * non-zero dl_time. They might have already been broken
6876                          * and could be processed by the laundromat outside of
6877                          * the state_lock. Just leave them be.
6878                          */
6879                         if (dp->dl_time != 0)
6880                                 continue;
6881
6882                         atomic_inc(&clp->cl_refcount);
6883                         WARN_ON(!unhash_delegation_locked(dp));
6884                         list_add(&dp->dl_recall_lru, victims);
6885                 }
6886                 ++count;
6887                 /*
6888                  * Despite the fact that these functions deal with
6889                  * 64-bit integers for "count", we must ensure that
6890                  * it doesn't blow up the clp->cl_refcount. Throw a
6891                  * warning if we start to approach INT_MAX here.
6892                  */
6893                 WARN_ON_ONCE(count == (INT_MAX / 2));
6894                 if (count == max)
6895                         break;
6896         }
6897         spin_unlock(&state_lock);
6898         return count;
6899 }
6900
6901 static u64
6902 nfsd_print_client_delegations(struct nfs4_client *clp)
6903 {
6904         u64 count = nfsd_find_all_delegations(clp, 0, NULL);
6905
6906         nfsd_print_count(clp, count, "delegations");
6907         return count;
6908 }
6909
6910 u64
6911 nfsd_inject_print_delegations(void)
6912 {
6913         struct nfs4_client *clp;
6914         u64 count = 0;
6915         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6916                                                 nfsd_net_id);
6917
6918         if (!nfsd_netns_ready(nn))
6919                 return 0;
6920
6921         spin_lock(&nn->client_lock);
6922         list_for_each_entry(clp, &nn->client_lru, cl_lru)
6923                 count += nfsd_print_client_delegations(clp);
6924         spin_unlock(&nn->client_lock);
6925
6926         return count;
6927 }
6928
6929 static void
6930 nfsd_forget_delegations(struct list_head *reaplist)
6931 {
6932         struct nfs4_client *clp;
6933         struct nfs4_delegation *dp, *next;
6934
6935         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6936                 list_del_init(&dp->dl_recall_lru);
6937                 clp = dp->dl_stid.sc_client;
6938                 revoke_delegation(dp);
6939                 put_client(clp);
6940         }
6941 }
6942
6943 u64
6944 nfsd_inject_forget_client_delegations(struct sockaddr_storage *addr,
6945                                       size_t addr_size)
6946 {
6947         u64 count = 0;
6948         struct nfs4_client *clp;
6949         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6950                                                 nfsd_net_id);
6951         LIST_HEAD(reaplist);
6952
6953         if (!nfsd_netns_ready(nn))
6954                 return count;
6955
6956         spin_lock(&nn->client_lock);
6957         clp = nfsd_find_client(addr, addr_size);
6958         if (clp)
6959                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
6960         spin_unlock(&nn->client_lock);
6961
6962         nfsd_forget_delegations(&reaplist);
6963         return count;
6964 }
6965
6966 u64
6967 nfsd_inject_forget_delegations(u64 max)
6968 {
6969         u64 count = 0;
6970         struct nfs4_client *clp;
6971         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
6972                                                 nfsd_net_id);
6973         LIST_HEAD(reaplist);
6974
6975         if (!nfsd_netns_ready(nn))
6976                 return count;
6977
6978         spin_lock(&nn->client_lock);
6979         list_for_each_entry(clp, &nn->client_lru, cl_lru) {
6980                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
6981                 if (max != 0 && count >= max)
6982                         break;
6983         }
6984         spin_unlock(&nn->client_lock);
6985         nfsd_forget_delegations(&reaplist);
6986         return count;
6987 }
6988
6989 static void
6990 nfsd_recall_delegations(struct list_head *reaplist)
6991 {
6992         struct nfs4_client *clp;
6993         struct nfs4_delegation *dp, *next;
6994
6995         list_for_each_entry_safe(dp, next, reaplist, dl_recall_lru) {
6996                 list_del_init(&dp->dl_recall_lru);
6997                 clp = dp->dl_stid.sc_client;
6998                 /*
6999                  * We skipped all entries that had a zero dl_time before,
7000                  * so we can now reset the dl_time back to 0. If a delegation
7001                  * break comes in now, then it won't make any difference since
7002                  * we're recalling it either way.
7003                  */
7004                 spin_lock(&state_lock);
7005                 dp->dl_time = 0;
7006                 spin_unlock(&state_lock);
7007                 nfsd_break_one_deleg(dp);
7008                 put_client(clp);
7009         }
7010 }
7011
7012 u64
7013 nfsd_inject_recall_client_delegations(struct sockaddr_storage *addr,
7014                                       size_t addr_size)
7015 {
7016         u64 count = 0;
7017         struct nfs4_client *clp;
7018         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7019                                                 nfsd_net_id);
7020         LIST_HEAD(reaplist);
7021
7022         if (!nfsd_netns_ready(nn))
7023                 return count;
7024
7025         spin_lock(&nn->client_lock);
7026         clp = nfsd_find_client(addr, addr_size);
7027         if (clp)
7028                 count = nfsd_find_all_delegations(clp, 0, &reaplist);
7029         spin_unlock(&nn->client_lock);
7030
7031         nfsd_recall_delegations(&reaplist);
7032         return count;
7033 }
7034
7035 u64
7036 nfsd_inject_recall_delegations(u64 max)
7037 {
7038         u64 count = 0;
7039         struct nfs4_client *clp, *next;
7040         struct nfsd_net *nn = net_generic(current->nsproxy->net_ns,
7041                                                 nfsd_net_id);
7042         LIST_HEAD(reaplist);
7043
7044         if (!nfsd_netns_ready(nn))
7045                 return count;
7046
7047         spin_lock(&nn->client_lock);
7048         list_for_each_entry_safe(clp, next, &nn->client_lru, cl_lru) {
7049                 count += nfsd_find_all_delegations(clp, max - count, &reaplist);
7050                 if (max != 0 && ++count >= max)
7051                         break;
7052         }
7053         spin_unlock(&nn->client_lock);
7054         nfsd_recall_delegations(&reaplist);
7055         return count;
7056 }
7057 #endif /* CONFIG_NFSD_FAULT_INJECTION */
7058
7059 /*
7060  * Since the lifetime of a delegation isn't limited to that of an open, a
7061  * client may quite reasonably hang on to a delegation as long as it has
7062  * the inode cached.  This becomes an obvious problem the first time a
7063  * client's inode cache approaches the size of the server's total memory.
7064  *
7065  * For now we avoid this problem by imposing a hard limit on the number
7066  * of delegations, which varies according to the server's memory size.
7067  */
7068 static void
7069 set_max_delegations(void)
7070 {
7071         /*
7072          * Allow at most 4 delegations per megabyte of RAM.  Quick
7073          * estimates suggest that in the worst case (where every delegation
7074          * is for a different inode), a delegation could take about 1.5K,
7075          * giving a worst case usage of about 6% of memory.
7076          */
7077         max_delegations = nr_free_buffer_pages() >> (20 - 2 - PAGE_SHIFT);
7078 }
7079
7080 static int nfs4_state_create_net(struct net *net)
7081 {
7082         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7083         int i;
7084
7085         nn->conf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7086                         CLIENT_HASH_SIZE, GFP_KERNEL);
7087         if (!nn->conf_id_hashtbl)
7088                 goto err;
7089         nn->unconf_id_hashtbl = kmalloc(sizeof(struct list_head) *
7090                         CLIENT_HASH_SIZE, GFP_KERNEL);
7091         if (!nn->unconf_id_hashtbl)
7092                 goto err_unconf_id;
7093         nn->sessionid_hashtbl = kmalloc(sizeof(struct list_head) *
7094                         SESSION_HASH_SIZE, GFP_KERNEL);
7095         if (!nn->sessionid_hashtbl)
7096                 goto err_sessionid;
7097
7098         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7099                 INIT_LIST_HEAD(&nn->conf_id_hashtbl[i]);
7100                 INIT_LIST_HEAD(&nn->unconf_id_hashtbl[i]);
7101         }
7102         for (i = 0; i < SESSION_HASH_SIZE; i++)
7103                 INIT_LIST_HEAD(&nn->sessionid_hashtbl[i]);
7104         nn->conf_name_tree = RB_ROOT;
7105         nn->unconf_name_tree = RB_ROOT;
7106         nn->boot_time = get_seconds();
7107         nn->grace_ended = false;
7108         nn->nfsd4_manager.block_opens = true;
7109         INIT_LIST_HEAD(&nn->nfsd4_manager.list);
7110         INIT_LIST_HEAD(&nn->client_lru);
7111         INIT_LIST_HEAD(&nn->close_lru);
7112         INIT_LIST_HEAD(&nn->del_recall_lru);
7113         spin_lock_init(&nn->client_lock);
7114
7115         spin_lock_init(&nn->blocked_locks_lock);
7116         INIT_LIST_HEAD(&nn->blocked_locks_lru);
7117
7118         INIT_DELAYED_WORK(&nn->laundromat_work, laundromat_main);
7119         get_net(net);
7120
7121         return 0;
7122
7123 err_sessionid:
7124         kfree(nn->unconf_id_hashtbl);
7125 err_unconf_id:
7126         kfree(nn->conf_id_hashtbl);
7127 err:
7128         return -ENOMEM;
7129 }
7130
7131 static void
7132 nfs4_state_destroy_net(struct net *net)
7133 {
7134         int i;
7135         struct nfs4_client *clp = NULL;
7136         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7137
7138         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7139                 while (!list_empty(&nn->conf_id_hashtbl[i])) {
7140                         clp = list_entry(nn->conf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7141                         destroy_client(clp);
7142                 }
7143         }
7144
7145         for (i = 0; i < CLIENT_HASH_SIZE; i++) {
7146                 while (!list_empty(&nn->unconf_id_hashtbl[i])) {
7147                         clp = list_entry(nn->unconf_id_hashtbl[i].next, struct nfs4_client, cl_idhash);
7148                         destroy_client(clp);
7149                 }
7150         }
7151
7152         kfree(nn->sessionid_hashtbl);
7153         kfree(nn->unconf_id_hashtbl);
7154         kfree(nn->conf_id_hashtbl);
7155         put_net(net);
7156 }
7157
7158 int
7159 nfs4_state_start_net(struct net *net)
7160 {
7161         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7162         int ret;
7163
7164         ret = nfs4_state_create_net(net);
7165         if (ret)
7166                 return ret;
7167         locks_start_grace(net, &nn->nfsd4_manager);
7168         nfsd4_client_tracking_init(net);
7169         printk(KERN_INFO "NFSD: starting %ld-second grace period (net %x)\n",
7170                nn->nfsd4_grace, net->ns.inum);
7171         queue_delayed_work(laundry_wq, &nn->laundromat_work, nn->nfsd4_grace * HZ);
7172         return 0;
7173 }
7174
7175 /* initialization to perform when the nfsd service is started: */
7176
7177 int
7178 nfs4_state_start(void)
7179 {
7180         int ret;
7181
7182         ret = set_callback_cred();
7183         if (ret)
7184                 return ret;
7185
7186         laundry_wq = alloc_workqueue("%s", WQ_UNBOUND, 0, "nfsd4");
7187         if (laundry_wq == NULL) {
7188                 ret = -ENOMEM;
7189                 goto out_cleanup_cred;
7190         }
7191         ret = nfsd4_create_callback_queue();
7192         if (ret)
7193                 goto out_free_laundry;
7194
7195         set_max_delegations();
7196         return 0;
7197
7198 out_free_laundry:
7199         destroy_workqueue(laundry_wq);
7200 out_cleanup_cred:
7201         cleanup_callback_cred();
7202         return ret;
7203 }
7204
7205 void
7206 nfs4_state_shutdown_net(struct net *net)
7207 {
7208         struct nfs4_delegation *dp = NULL;
7209         struct list_head *pos, *next, reaplist;
7210         struct nfsd_net *nn = net_generic(net, nfsd_net_id);
7211         struct nfsd4_blocked_lock *nbl;
7212
7213         cancel_delayed_work_sync(&nn->laundromat_work);
7214         locks_end_grace(&nn->nfsd4_manager);
7215
7216         INIT_LIST_HEAD(&reaplist);
7217         spin_lock(&state_lock);
7218         list_for_each_safe(pos, next, &nn->del_recall_lru) {
7219                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7220                 WARN_ON(!unhash_delegation_locked(dp));
7221                 list_add(&dp->dl_recall_lru, &reaplist);
7222         }
7223         spin_unlock(&state_lock);
7224         list_for_each_safe(pos, next, &reaplist) {
7225                 dp = list_entry (pos, struct nfs4_delegation, dl_recall_lru);
7226                 list_del_init(&dp->dl_recall_lru);
7227                 put_clnt_odstate(dp->dl_clnt_odstate);
7228                 nfs4_put_deleg_lease(dp->dl_stid.sc_file);
7229                 nfs4_put_stid(&dp->dl_stid);
7230         }
7231
7232         BUG_ON(!list_empty(&reaplist));
7233         spin_lock(&nn->blocked_locks_lock);
7234         while (!list_empty(&nn->blocked_locks_lru)) {
7235                 nbl = list_first_entry(&nn->blocked_locks_lru,
7236                                         struct nfsd4_blocked_lock, nbl_lru);
7237                 list_move(&nbl->nbl_lru, &reaplist);
7238                 list_del_init(&nbl->nbl_list);
7239         }
7240         spin_unlock(&nn->blocked_locks_lock);
7241
7242         while (!list_empty(&reaplist)) {
7243                 nbl = list_first_entry(&reaplist,
7244                                         struct nfsd4_blocked_lock, nbl_lru);
7245                 list_del_init(&nbl->nbl_lru);
7246                 posix_unblock_lock(&nbl->nbl_lock);
7247                 free_blocked_lock(nbl);
7248         }
7249
7250         nfsd4_client_tracking_exit(net);
7251         nfs4_state_destroy_net(net);
7252 }
7253
7254 void
7255 nfs4_state_shutdown(void)
7256 {
7257         destroy_workqueue(laundry_wq);
7258         nfsd4_destroy_callback_queue();
7259         cleanup_callback_cred();
7260 }
7261
7262 static void
7263 get_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7264 {
7265         if (HAS_STATE_ID(cstate, CURRENT_STATE_ID_FLAG) && CURRENT_STATEID(stateid))
7266                 memcpy(stateid, &cstate->current_stateid, sizeof(stateid_t));
7267 }
7268
7269 static void
7270 put_stateid(struct nfsd4_compound_state *cstate, stateid_t *stateid)
7271 {
7272         if (cstate->minorversion) {
7273                 memcpy(&cstate->current_stateid, stateid, sizeof(stateid_t));
7274                 SET_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7275         }
7276 }
7277
7278 void
7279 clear_current_stateid(struct nfsd4_compound_state *cstate)
7280 {
7281         CLEAR_STATE_ID(cstate, CURRENT_STATE_ID_FLAG);
7282 }
7283
7284 /*
7285  * functions to set current state id
7286  */
7287 void
7288 nfsd4_set_opendowngradestateid(struct nfsd4_compound_state *cstate,
7289                 union nfsd4_op_u *u)
7290 {
7291         put_stateid(cstate, &u->open_downgrade.od_stateid);
7292 }
7293
7294 void
7295 nfsd4_set_openstateid(struct nfsd4_compound_state *cstate,
7296                 union nfsd4_op_u *u)
7297 {
7298         put_stateid(cstate, &u->open.op_stateid);
7299 }
7300
7301 void
7302 nfsd4_set_closestateid(struct nfsd4_compound_state *cstate,
7303                 union nfsd4_op_u *u)
7304 {
7305         put_stateid(cstate, &u->close.cl_stateid);
7306 }
7307
7308 void
7309 nfsd4_set_lockstateid(struct nfsd4_compound_state *cstate,
7310                 union nfsd4_op_u *u)
7311 {
7312         put_stateid(cstate, &u->lock.lk_resp_stateid);
7313 }
7314
7315 /*
7316  * functions to consume current state id
7317  */
7318
7319 void
7320 nfsd4_get_opendowngradestateid(struct nfsd4_compound_state *cstate,
7321                 union nfsd4_op_u *u)
7322 {
7323         get_stateid(cstate, &u->open_downgrade.od_stateid);
7324 }
7325
7326 void
7327 nfsd4_get_delegreturnstateid(struct nfsd4_compound_state *cstate,
7328                 union nfsd4_op_u *u)
7329 {
7330         get_stateid(cstate, &u->delegreturn.dr_stateid);
7331 }
7332
7333 void
7334 nfsd4_get_freestateid(struct nfsd4_compound_state *cstate,
7335                 union nfsd4_op_u *u)
7336 {
7337         get_stateid(cstate, &u->free_stateid.fr_stateid);
7338 }
7339
7340 void
7341 nfsd4_get_setattrstateid(struct nfsd4_compound_state *cstate,
7342                 union nfsd4_op_u *u)
7343 {
7344         get_stateid(cstate, &u->setattr.sa_stateid);
7345 }
7346
7347 void
7348 nfsd4_get_closestateid(struct nfsd4_compound_state *cstate,
7349                 union nfsd4_op_u *u)
7350 {
7351         get_stateid(cstate, &u->close.cl_stateid);
7352 }
7353
7354 void
7355 nfsd4_get_lockustateid(struct nfsd4_compound_state *cstate,
7356                 union nfsd4_op_u *u)
7357 {
7358         get_stateid(cstate, &u->locku.lu_stateid);
7359 }
7360
7361 void
7362 nfsd4_get_readstateid(struct nfsd4_compound_state *cstate,
7363                 union nfsd4_op_u *u)
7364 {
7365         get_stateid(cstate, &u->read.rd_stateid);
7366 }
7367
7368 void
7369 nfsd4_get_writestateid(struct nfsd4_compound_state *cstate,
7370                 union nfsd4_op_u *u)
7371 {
7372         get_stateid(cstate, &u->write.wr_stateid);
7373 }